Anda di halaman 1dari 456


14-17 October
Timisoara, ROMANIA
Edited by: Bebo White,
Pedro Isaas
and Diana Andone
international association for development of the information society





OCTOBER 14-17, 2010

Organised by
International Association for Development of the Information Society

Co-Organised by


Copyright 2010
All rights reserved
This work is subject to copyright. All rights are reserved, whether the whole or part of the material
is concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other way, and storage in data banks.
Permission for use must always be obtained from IADIS Press. Please contact

Edited by Bebo White, Pedro Isaas and Diana Andone

Associate Editors: Lus Rodrigues and Patrcia Barbosa

ISBN: 978-972-8939-25-0




Daniel Avila Vecchiato, Maria Beatriz Felgar de Toledo, Marcelo Fantinato and
Itana Maria de Souza Gimenes
Niklas Eriksson and Peter Strandvik
Cldio Martins, Joo Rosa, Laerte Franco and Jorge Barbosa
Herli J. de Menezes, Sean W. M. Siqueira and Leila Cristina V. de Andrade
Douglas O. de Freitas, Marcos A. R. Silva, Johana M. R. Villena, Bruno A. Sugiyama,
Gilberto Astolfi and Junia C. Anacleto
Hideaki Goto
Roland Grlitz, Benjamin Seip, Asarnusch Rashid and Valentin Zacharias
Ruth Diko Wario and Theo McDonald
Malek Barhoush and J. William Atwood

Gregory Triantafillidis and John Garofalakis
Rita Oluchi Orji, Yasemin Yardimci Cetin and Sevgi Ozkan
Paulo Caetano da Silva, Marcelo Mendona dos Santos and Valria Cesrio Times
Takahiro Kawamura, I Shin, Hiroyuki Nakagawa, Yasuyuki Tahara and Akihiko Ohsuga
Claudia Iacob and Li Zhu
Rafael Ferreira, Rinaldo J. Lima, Ig Ibert Bittencourt, Dimas Melo Filho, Olavo Holanda,
Evandro Costa, Fred Freitas and Ldia Melo
Juliana de Melo Bezerra and Celso Massaki Hirata
Diogo Arantes Fonseca Gonalves da Cunha, Artur Ferreira da Silva and Jos Figueiredo
Roberto Pereira, M. Ceclia C. Baranauskas, Sergio Roberto P. da Silva, Jos Valderlei da Silva
and Filipe Roseiro Cgo
Shohei Yokoyama and Hiroshi Ishikawa
Alssio Miranda Jnior and Laura S. Garca
Ghada Aldehim, Pam Mayhew and Majed Alshamari
Gerald Hbsch, Christian Liebing, Josef Spillner and Alexander Schill
Lara Schibelsky Godoy Piccolo and Maria Ceclia Calani Baranauskas
Daniel Kappler and Paul Darbyshire
Mauro Andreolini, Sara Casolari and Stefania Tosi
Priscila Doria and Marco Aurlio Spohn
Sidney Doria and Marco Aurlio Spohn
Michal Barla and Mria Bielikov
Christoph Trattner
Chyng-Yang Jang and Michael A. Stefanone


Keiichi Shiraishi, Yoshiro Imai and Yukio Hori
Hale Ilgaz and Petek Akar
Cayley Guimares, Diego R. Antunes, Daniela F. G. Trindade, Rafaella A. L. Silva,
Sueli Fernandes, Alssio M. Jr. and Laura Snchez Garca
Luca Dan Serbanati and Andrei Vasilateanu
Wangsen Feng, Bei Zhang, Yunni Xia, Jiajing Li and Tao Meng
Anne Kmpel, Iris Braun, Josef Spillner and Alexander Schill
Vinicius Miana Bezerra and Selma Melnikoff
Claudia Dossena and Alberto Francesconi
George Gkotsis, Nikos Karacapilidis, Costas Pappis and Nikos Tsirakis
Yasemin Koak Usluel, Pnar Nuholu and Bahadr Yildiz
Mrcia R. M. Martinez, Sueli A. Varani, Edgar L. Banhesse and Clenio F. Salviano
Roberto Paiano and Andrea Pandurino
Marius Feldmann, Felix Martens, Georg Berndt, Josef Spillner and Alexander Schill
Dieter Blomme, Heiko Desruelle, Frank Gielen and Filip De Turck
Michael Barros, Anderson Costa and Reinaldo Gomes
Lukas Gerhold, Hilal Tekoglu, Markus Dorn and Michael Binder
Julia Kasmire and Alfredas Chmieliauskas
Sotirios Karetsos and Dias Haralampopoulos
Miranda, Alssio; Paula, ; Garca, L.S; Capovilla, F.C; Direne, A.I; Sunye, M.S.;
Castilho, M.A.; Bona, L.E; Silva, F.; Machado, D.; Duarte, J.; Bueno, J.
Raghad Jawad Ahmed
Juliana Bueno, Laura Snchez Garca and Alssio Miranda Jnior


Cayley Guimares, Diego R. Antunes, Alice G. de Paula and Alssio Miranda Jr
Sergio R. Coria
Lara Cini, Matthew Montebello and Vanessa Camilleri
Francesco De Angelis, Roberto Gagliardi, Alberto Polzonetti and Barbara Re
Iasmina Ermalai and Radu Vasiu


Tiago Frana Melo de Lima and Cludio Gouvea dos Santos
Silviu Panica, Marian Neagul and Dana Petcu
Yukie Majima, Masato Soga and Yasuko Maekawa
Yukie Majima, Yumiko Nakamura, Yasuko Maekawa, Hiroko Makino, Yukari Nakajima and
Mizuko Hiramatsu
Matthias Heinrich and Antje Boehm-Peters
Matthew Sammut, Matthew Montebello and Vanessa Camilleri
Angela M. Alves, Giancarlo Stefanuto, Paula F. D. Castro and Marcelo Pessa
Claudio Gottschalg Duque and Mauricio Rocha Lyra

Bernd Stadlhofer and Peter Salhofer
Adelaide Bianchini



These proceedings contain the papers and posters of the IADIS International Conference
WWW/Internet 2010, which was organised by the International Association for
Development of the Information Society and co-organised by "Politehnica" University of
Timisoara, Romania, 14-17 October 2010.

The IADIS WWW/Internet 2010 Conference aims to address the main issues of concern
within WWW/Internet. WWW and Internet had a huge development in recent years.
Aspects of concern are no longer just technical anymore but other aspects have arisen. This
conference aims to cover both technological as well as non-technological issues related to
these developments

Submissions were accepted under the following main tracks and topics:
Web 2.0
Collaborative Systems
Social Networks
Enterprise Wikis and Blogging
Mashups and Web Programming
Tagging and User Rating Systems
Citizen J ournalism

Semantic Web and XML
Semantics Web Architectures
Semantic Web Middleware
Semantic Web Services
Semantic Web Agents
Applications of Semantic Web
Semantic Web Data Management
Information Retrieval in Semantic Web

Applications and Uses
e-Commerce / e-Business
Digital Libraries
Web Services/SaaS
Application Interoperability
Web-based multimedia technologies

Services, Architectures and Web Development
Wireless Web
Mobile Web
Cloud/Grid Computing
Web Metrics
Web Standards
Internet Architectures
Network Algorithms
Network Architectures
Network Computing
Network Management
Network Performance
Content Delivery Technologies
Protocols and Standards
Traffic Models
Research Issues
Web Science
Digital Rights Management
Human Computer Interaction and Usability
Web Security and Privacy
Online Trust and Reputation Systems
Data Mining
Information Retrieval
Search Engine Optimization
The IADIS WWW/Internet 2010 Conference had 153 submissions from more than 28
countries. Each submission has been anonymously reviewed by an average of four
independent reviewers, to ensure the final high standard of the accepted submissions. The
final result was the approval of 30 full papers, which means that the acceptance rate was
below 20%. A few more papers have been accepted as short papers, reflection papers,
posters and doctoral consortium. Best papers will be selected for publishing as extended
versions in the IADIS International J ournal on WWW/Internet (IJ WI) and in other selected
The conference, besides the presentation of full papers, short papers, reflection papers,
posters and doctoral consortium presentations also included two keynote presentations from
internationally distinguished researchers. We would like also to express our gratitude to
Professor Mark Frydenberg, Senior Lecturer, Bentley University, USA and Dr. Molly
Holzschlag, Web Evangelist, Developer Relations, Opera Software, USA, for being our
keynote speakers.
As we all know, organising a conference requires the effort of many individuals. We would
like to thank all members of the Program Committee for their hard work in reviewing and
selecting the papers that appear in the proceedings.


We are especially grateful to the authors who submitted their papers to this conference and
to the presenters who provided the substance of the meeting
These Proceeding book contain a rich experience of the academic & research institutions
and the industry on diverse themes related to the Internet and Web. We do hope that
researchers, knowledge workers and innovators both in academia and the industry will find
it a valuable reference material.
Last but not the least, we hope that everybody will have a good time in Timisoara and we
invite all participants for the next years edition of the IADIS International Conference
WWW/Internet 2011.

Pedro Isaas, Universidade Aberta (Portuguese Open University), Portugal
Diana Andone, "Politehnica" University of Timisoara, Romania
Conference Co-Chairs

Bebo White, Stanford University, USA
Program Chair

Timisoara, Romania
14 October 2010



Pedro Isaas, Universidade Aberta (Portuguese Open University), Portugal
Diana Andone, "Politehnica" University of Timisoara, Romania

Bebo White, Stanford University, USA


Abdolhossein Sarrafzadeh, UNITEC New Zealand, New Zealand
Adrian Stoica, University Of Patras, Greece
Alberto Barrn-Cedeo, Universidad Politcnica Valencia, Spain
Alexander Krner, Dfki, Germany
Alexander Pokahr, Pokahr, Germany
Alexander Schill, TU Dresden, Germany
Alexander Troussov, IBM Dublin Center for Advanced Studies, Ireland
Alexander Whrer, University Of Vienna, Austria
Amali Weerasinghe, University Of Canterbury, New Zealand
Ana Sanz, Universidad Carlos III De Madrid, Spain
Ana Regina Cavalcanti Da Rocha, Centroin, Brazil
Ananya S. Guha, Indira Gandhi National Open University, India
Andrea Kienle, University of Applied Sciences, Dortmund, Germany
Andreas Papasalouros, University Of The Aegean, Greece
Andreas Schrader, ISNM - International School Of New Media at the Un, Germany
Anglica Antonio, Universidad Politcnica De Madrid, Spain
Angelo Di Iorio, University Of Bologna, Italy
Anirban Kundu, West Bengal University Of Technology, India
Antonio Reyes, Universidad Politcnica De Valencia, Spain
Arturo Mora-Soto, Carlos III University Of Madrid, Spain
Asadullah Shaikh, Universitat Oberta de Catalunya, Spain
Batrice Rumpler, Insa Of Lyon, France
Beob Kyun Kim, Korea Institute of Science and Technology Informat, South Korea
Bianca Shoen, University College Dublin, Ireland
Bo Hu, Universitt Der Bundeswehr Mnchen, Germany
Brahmananda Sapkota, University Of Twente, Netherlands
Cao Rong Zeng, IBM China Research Lab, China
Carsten Ulrich, Shanghai J iao Tong University, China
Chih Cheng Hung, Southern Polytechnic State University, Usa
Christos Bouras, University Of Patras, Greece
Christos Georgiadis, University Of Macedonia Thessaloniki, Greece
Chunhua Tian, IBM China Research Lab, China


Ciprian Dobre, University Politehnica Of Bucharest, Romania
Clodoveu Davis, Universidade Federal de Minas Gerais, Brazil
Constantina Costopoulou, Agricultural University Of Athens, Greece
Costas Yialouris, Agricultural University Of Athens, Greece
Daniel Cunliffe, University Of Glamorgan, United Kingdom
Daniel Schuster, TU Dresden, Germany
Debajyoti Mukhopadhyay, Calcutta Business School, India
Demetrios Sampson, University Of Piraeus, Greece
Diana Prez Marn, Universidad Rey J uan Carlos, Spain
Dirk Thissen, Rwth Aachen, Germany
Dongqiang Yang, Flinders University, Australia
Elena Calude, Massey University At Albany, New Zealand
Elena Camossi, European Commission, J oint Research Centre, Italy
Elias Kirche, Florida Gulf Coast University, USA
Elif Ubeyli, Tobb Economics And Technology University, Turkey
Eloisa Vargiu, DIEE, Italy
Eva Sderstrm, University Of Skvde, Sweden
Ezendu Ariwa, London Metropolitan University, United Kingdom
Fahim Akhter, Zayed University, United Arab Emirates
Fan Zhao, Florida Gulf Coast University, Usa
Fatos Xhafa, Universitat Politcnica De Catalunya, Spain
Florence Sedes, Universite Paul Sabatier Of Toulouse, France
Florin Pop, University Politehnica Of Bucharest, Romania
Francesco Guerra, Universita Di Modena E Reggio Emilia, Italy
Francesco Pagliarecci, Polytechnical University Of Marche, Italy
Fuensanta Medina-Domnguez, Carlos III Technical University of Madrid, Spain
Fuhua Lin, Athabasca University, Canada
Geneen Stubbs, University Of Glamorgan, United Kingdom
George Gkotsis, University Of Patras, Greece
George Koutromanos, University Of Athens, Greece
George Vouros, University Of The Aegean, Greece
Gheorghe Cosmin Silaghi, Babes-Bolyai University of Cluj-Napoca, Romania
Giuseppe Patane, Cnr-imati, Italy
Guido Roessling, Darmstadt University Of Technology, Germany
Gustavo Rossi, Universidad Nacional De La Plata, Argentina
Hakikur Rahman, Institute of Computer Management & Science (ICMS), Bangladesh
Hamid Mcheick, Universit Du Qubec Chicoutimi, Canada
Henry Oinas-Kukkonen, University Of Oulu, Finland, Finland
Holger Hinrichs, University Of Applied Sciences Lbeck, Germany
Horst Hellbrck, University Of Applied Sciences Lbeck Horst, Germany
Idris Rai, Lancaster University, United Kingdom
Ioan Toma, University Innsbruck, Austria
J . Enrique Agudo, Universidad De Extremadura, Spain
J . K. Vijayakumar, King Abdullah University of Science and Technology, Saudi Arabia
J aime Ramrez Ramrez, Universidad Politcnica De Madrid, Spain
J ames Thong, Hong Kong University Of Science And Technology, Hong Kong
J ames Walden, Northern Kentucky University, United States
J an Zizka, SoNet/Dept. of Informatics, FBE, Mendel University, Czech Republic
J anez Brank, J ozef Stefan Institute, Slovenia

J avier Saldaa, Universidad Carlos Iii De Madrid, Spain
J essica Rubart, Arvato Services, Germany
J injun Chen, Swinburne University Of Technology, Australia
J ohn Garofalakis, University of Patras, Greece
J rg Roth, University Of Applied Sciences Nuremberg, Germany
J os Laurindo Campos dos Santos, Instituto Nacional De Pesquisas Na Amazonia, Brazil
J uan Ramn Prez Prez, Universidad De Oviedo, Spain
J uhnyoung Lee, IBM T. J . Watson Research Center, Usa
J ulie Yazici, Florida Gulf Coast University, Usa
Kai J akobs, RWTH Aachen University, Germany
Kangsoo You, J eonju University, Korea (south)
Konstantinos Kotis, University Of The Aegean, Greece
Krzysztof Walkowiak, Wroclaw University Of Technology, Poland
Lars Braubach, Informatik Der Universitt Hamburg, Germany
Lazaros Iliadis, Democritus University Of Thrace, Greece
Leila Shafti, Universidad Autonoma De Madrid, Spain
Liping Liu, University Of Akron, Usa
Luca Spalazzi, Polytechnic University Of Marche, Italy
Lucia Rapanotti, The Open University, United Kingdom
Manolis Tzagarakis, University Of Patras , Greece
Manuel Montes y Gmez, INAOE, Mexico
Mara Nikolaidou, Harokopio University Of Athens, Greece
Margherita Antona, ICS - FORTH, Greece
Maria Claudia Buzzi, CNR - IIT, Italy
Markos Hatzitaskos, University Of Patras, Greece
Martin Gaedke, Chemnitz University Of Technology, Germany
Massimo Marchiori, Unipd/ Utilabs, Italy
Mauricio Marin, Yahoo! Research Santiago, Chile
Maytham Safar, Kuwait University, Kuwait
Michael Mrissa, University of Lyon, France
Michal Wozniak, Wroclaw University Of Technology, Poland
Michalis Vaitis, University of the Aegean, Greece
Miguel-Angel Sicilia Urbn, University Of Alcal, Spain
Ming-Puu Chen, National Taiwan Normal University, Taiwan
Miroslav Bures, Czech Technical University In Prague, Czech Republic
Nikolaos Tselios, University Of Patras, Greece
Nikos Karacapilidis, University Of Patras, Greece
Nikos Karousos, University Of Patras, Greece
Nikos Tsirakis, University Of Patras, Greece
Olga Santos Martn, Uned, Spain
Omair Shafiq, University Of Calgary, Canada
Ourania Hatzi, Harokopio University Of Athens, Greece
Paolo Rosso, Universidad Politcnica De Valencia, Spain
Pedro Paredes, Universidad Autnoma De Madrid, Spain
Penelope Markellou, University Of Patras, Greece
Peter Geczy, AIST (National Institute of Advanced Industrial Sc, J apan
Raquel Hervas, Universidad Complutense de Madrid, Spain
Robert Burduk, Wroclaw University Of Technology, Poland
Robert Goodwin, Flinders University, Australia

Runa J esmin, University Of London, United Kingdom
Samad Kolahi, Unitec Institute Of Technology, New Zealand
Sattanathan Subramanian, Uni-BCCS, Norway
Silvia Gaiani, Universit Di Bologna, Italy
Sotiris Karetsos, Agricultural University Of Athens, Greece
Stefan Dietze, The Open University, United Kingdom
Stefan Fischer, Universitt Zu Lbeck, Germany
Steven Demurjian, The University Of Connecticut, USA
Stuart Cunningham, Glyndwr University, United Kingdom
Sung-kook Han, Wonkwang University, Korea, Republic Of
Tayana Conte, Federal University Of Rio De J aneiro, Brazil
Tharrenos Bratitsis, University Of Western Macedonia, Greece
Thomas Springer, Tu Dresden, Germany
Tobias Buerger, Salzburg Research, Austria
Tomoo Inoue, University Of Tsukuba, J apan
Tukaram Fulzele, Indira Gandhi National Open University, India
Vassilis Kapsalis, Industrial Systems Institute, Greece
Vic Grout, University Of Wales, United Kingdom
Wei Hao, Northern Kentucky University, United States
Wilhelm Schwieren, Rwth Aachen University, Germany
Yuanbo Guo, Microsoft, USA



by Mark Frydenberg, Senior Lecturer
Bentley University, USA


Web 2.0 marked the evolution of the World Wide Web from a one way web filled with static
content to a dynamic read/write Web that has become a platform for applications promoting
collaboration and communication, linking people as well as the digital information they share. As
society creates more online content than it can consume, the need to organize, search, and
understand Web content becomes critical. Tim OReillys described Internet of Things and Tim
Berners-Lees vision of a Semantic Web are not far off. Or are they? This session will share
examples of emerging Web trends, technologies, and patterns and look at what is possible when
both data and applications live on the Web.

by Molly Holzschlag, Web Evangelist
Developer Relations, Opera Software, USA


In this keynote session, Ms. Holzschlag will discuss the OpenWeb vision, where it emerged from
and what it provides to real-world Web developers. A look at the advancements in browser
evolution as well as visual examples of the Open Web Technology Stack will demonstrate a
powerful, open-standards oriented approach to the evolving Web of applications.


Full Papers

Daniel Avila Vecchiato*, Maria Beatriz Felgar de Toledo*, Marcelo Fantinato**
and Itana Maria de Souza Gimenes***
*University of Campinas, Institute of Computing, Av. Albert Einstein, 1251, Campinas - SP, Brazil
**University of So Paulo, School of Arts, Sciences and Humanities, R. Arlindo Bttio, 1000, So Paulo - SP, Brazil
***University of Maring, Department of Informatics, Av. Colombo, 5.790, Maring - PR, Brazil
E-contracts usually describe inter-organizational business processes defining e-services to be provided and consumed as
well as non-functional requirements such as Quality of Service (QoS). Organizations involved in a cooperation need to
provide explicit guarantees in an e-contract, which can involve renegotiation of contractual clauses, penalty application or
intervention in the process execution. In this paper, feature modeling is used to represent e-services, QoS attributes and
control operations to be triggered when QoS attribute levels are not met. In addition, it describes FeatureContract, a
toolkit to support contract establishment based on features. An application example is exploited to show the proposed
approach feasibility. The main contributions are related to improved structure and reuse in information in e-contracts.
Business Process Management, e-contracts, contract negotiation, contract renegotiation, e-services.
The current Business Process Management (BPM) scenario includes: i) one or more organizations that
provide and/or consume electronic services (e-services); ii) negotiation and establishment of electronic
contracts (e-contracts), including quality of service (QoS) attributes and levels; iii) definition, enactment,
monitoring, and auditing of business process; and, iv) process analysis and optimization. E-contracts between
two or more partners interested in an inter-organizational business process establish the activities to be
performed and the obligations, permissions and rights related to each involved party. During contract
enactment, if a party is unable to fulfill contractual clauses, a contract renegotiation may be triggered.
Fantinato et al. (2010) have developed a new approach to reduce the complexity in the wide management
of e-contracts for Web services (WS-Contract). The approach, based on Software Product Line (PL)
concepts, is named PL4BPM (Product Line for Business Process Management). A predominant part of the
PL4BPM approach is the WS-Contract establishment process, focused on providing improved information
reuse and structuring based on the use of feature modeling (Fantinato et al. 2008).
This paper discusses the e-contract life cycle from negotiation, establishment, enactment to renegotiation
within the context of a feature-based BPM infrastructure. The main contributions are: i) an extension of a
contract metamodel to include, respectively, the control operations to be performed in case of e-contract
violation and represent it by using the WS-Agreement specification; ii) an extended infrastructure to support
e-contract negotiation and renegotiation, including the extension of the FeatureContract toolkit; iii) more
efficient management, and reuse of information necessary for the establishment and renegotiation of
e-contracts; and, iv) an example of the approach showing the various stages of an e-contract establishment.
The feature modeling technique used in the representation of e-services and QoS attributes has shown the
following advantages (Fantinato et al. 2008): flexibility for e-services specification; modularization facilities
for QoS attributes specification; and, structured representation of the optional and mandatory WS-contract
elements. This may also be advantageous in contract renegotiations as we will see.
The paper is organized as follows: Section 2 presents basic concepts; Section 3 presents some related
work; Section 4 presents the extended Feature and WS-Contract metamodels; Section 5 presents the
IADIS International Conference WWW/Internet 2010
extension made on the FeatureContract toolkit, within the proposed BPM infrastructure; Section 6 presents
an application example using the proposed approach; and, finally, Section 7 concludes the paper.
In this section, firstly e-contracts are described and then feature modeling concepts are presented as this is the
technique used in this approach to represent information to be included in e-contracts.
2.1 E-contracts
Organizations interested in Internet business partnerships must define details of the business process to be
enacted and express them in e-contracts (Fantinato et al. 2008). An e-contract defines details about the
organizations, the activities to be executed and the contractual clauses that must be met during process
enactment (Grefen et al. 2001). The clauses could be of three types: obligations, rights and prohibitions. The
obligation clauses include QoS of e-services within the inter-organizational process. In addition to the
functional aspect of e-contracts, there is also the legal aspect that will not be considered is this paper.
The activities to be contracted are e-services that may be implemented as Web services; when e-contracts
are called WS-Contracts. Web Services have spread as a promising technology for the effective automation
of inter-organizational interactions (Papazoglou 2007). The major benefit of it is the wide standardization
including: a language to describe service interfaces (WSDL), a service directory structure and APIs for
service publication and discovery (UDDI) and a communication protocol (SOAP).
The contractual clauses may be specified in languages such as WS-Agreement, a specification offering
assurance about Web Services execution (Andrieux et al. 2007). It was developed to represent and establish
agreements between two parts, usually a service provider and a consumer. The specification uses XML to
structure an agreement that is composed by elements such as Name, Context, Terms (Service Terms and
Guarantee Terms). WS-Agreement also allows the definition of service level objectives and qualifying
conditions associated with importance, penalties and rewards representing the business value to the provider.
2.2 Feature Modeling
Feature modeling captures and manages common points and variabilities in software product lines (Pohl et al.
2005). A feature model represents properties of some entity of interest. It can denote any functional or non-
functional property in the requirement, architectural or other levels. Features can be mandatory, optional or
alternative (Czarnecki et al. 2005). They are organized into a tree-like diagram in which a node represents a
feature and each feature can be described by a set of sub-features represented as descendant nodes. A feature
model describes a system family. A family member can be configured by selecting the desired features from
the model within the model variability limits, which is called feature model configuration.
The CrossFlow project is precursor in the area of inter-organizational business process (Grefen et al. 2001).
As some other earlier works, they use metamodels and templates to facilitate e-contract establishment. More
recently, Angelov and Grefen (2008b) defined an e-contract metamodel with different perspectives such as
function and communication perspective. They identify four e-contract lifecycle phases: i) information that
describes services and possible partners; ii) pre-contracting that customizes business operational aspects; iii)
contracting that establishes how the business process will be carried out; and iv) enactment that comprises the
execution of the contracted e-services. Angelov and Grefen (2008a) also defined a reference architecture to
contract systems development, using a component-based approach. It provides a component for each phase of
e-contracting (information, pre-contracting, contracting and enactment).
Bacarin et al. (2008) put forth a negotiation protocol with some primitive actions to assign property
values, to send offers, request for proposal (RFP) and votes. They identify the following phases: negotiation
ISBN: 978-972-8939-25-0 2010 IADIS
announcement, leader determination, objective announcement, negotiation setup, restriction announcement,
core negotiation, commit attempt, contract (re)construction.
Some works (Angelov & Grefen 2008b; Bacarin et al. 2008; Grefen et al. 2001) use e-contract template to
facilitate e-contract reuse. In a general way, the renegotiation issue is still not completely addressed in a
proper way by the works proposed in the literature. Some architectures and frameworks (Angelov & Grefen
2008a; Bacarin et al. 2008) allows contract update during process execution. However, none of them
specifies the actions to be performed in the case of contract violation. This work addresses the issue of
control operations to handle contract enactment in the case of any clause violation.
The BPM context involves business process composed by e-services. Our approach considers e-services
implemented as Web Services and hence the WS-Contracts to regulate the collaboration between the parties.
The WS-Contract is composed of: parties, e-services, contractual clauses and a business process. WS-BPEL
is used to define the parties and the orchestration of the e-services within an inter-organizational context
(Alves et al. 2007). E-services and QoS attributes are described in WSDL and WS-Agreement, respectively.
The WS-Contract metamodel, presented in Figure 1, represents the structure that supports the creation of
e-contracts. Originally, it was proposed by Fantinato et al. (2008) and was extended (in gray color) to
represent control operations. The metamodel comprises: i) Web Services described in the WSDL; ii)
Business process specified in the WS-BPEL; iii) QoS attributes of Web Services, described in
WS-Agreement; and, iv) Control operations to be handled in case of contract violation, also described in
The WS-Contract is composed of three sections, as follows:
WS-BPEL definitions: this section defines the business process using the terms Variable, Partner
Link and Activity (including Basic Activity and Structured Activity);
WSDL definitions: this is the services section, it contains the base elements Message Type, Partner
Link Type, Port Type and Operation. The last two describe the Web Services;
WS-Agreement definitions: it defines the business terms. The following elements are used to
define QoS levels: Service Property (including Variable) and Guarantee Term (including Service Scope,
Service Level Objective and Business Value List). It includes also:
- Control operations: represented within the WS-Agreement definitions, more specifically in the
Business Value List section using the penalty tags" as illustrated latter in Figure 3.
A useful strategy explored here is the use of contract templates. An e-contract template is defined only
once, but allows instantiation of distinct and similar contracts. To facilitate the creation of templates,
Fantinato et al. (2008) proposed a feature metamodel. It originally consisted of two sub-trees e-services and

Figure 1. WS-Contract metamodel
IADIS International Conference WWW/Internet 2010
QoS-attributes, but it has been extended to include a Control Operations sub-tree as presented in Figure 2.
The acronyms depicted in this figure are used to identify the feature type in the feature models, as in the
Figure 4 presented later. In Figure 2, each sub-tree is described as follows:
E-Services sub-tree: (mandatory) contains features representing the e-services offered by an
involved organization;
QoS-attributes sub-tree: (optional) contains features that represent the QoS attributes, which are
attached to e-services defined into the e-Services sub-tree. It includes choices of QoS attribute levels;
Control-operations sub-tree: (optional) it specifies control operations to be executed when QoS
attribute levels are not met. These attributes are attached to e-services defined into the e-Services sub-tree
and to QoS attributes defined into the QoS-attributes sub-tree. The Control Operation and Activity are
depicted latter in Figure 3, each control operation have a possibility to be controlled or not by an activity. The
element Value, in dark gray color, will only exist in the fine application activity and is used, to represent the
fine value.
The Control Operations sub-tree can be associated directly to an e-service or to specific QoS attributes.
The former is used as a default option whereas the latter is used as a specialization option. When a QoS
attribute is not met, if there are control operations settings defined for it, they are triggered; otherwise, if
there are control operations settings defined for the associated e-service, these control operations are
triggered instead. With this feature structure support, a unique set of control operations options, defined only
once, can be reused by all the QoS attributes and levels associated to all the e-services. During feature model
configuration, specific control operations options can be selected for each QoS attribute or for each e-service.
Figure 3 shows the possible Control Operations and the mapping with WS-Agreement section, as follows:
Renegotiation: used for contract update in three ways: i) Clause (QoS attribute) that can be
removed, added or updated. It can be necessary if new requirements in the inter-organization cooperation
appear; ii) Variable (QoS level) that can be renegotiated when triggering a penalty or control operations on a
process are not necessary; and, iii) Price that aservice or a QoS level price can be renegotiated. This can be
applied in services that are not having QoS attribute levels as expected;
Penalty Application: used to apply a penalty to the offending party. The penalty is a fine to
compensate some eventual loss. It can be selected, for example, if the QoS attribute availability" is not
fulfilled causing loss of money or clients;
Process: used to directly influence the business process execution. The available operations are: i)
Rollback that undoes operations already executed by the business process. It can be selected, for example,
for atomic e-services executed in a transactional way; ii) Suspend that stops the business process execution
until some condition is reached. It can be selected, for example, for the QoS attribute security, since the
process can be suspended until some security level is fulfilled; and, iii) Termination that terminates the
business process execution. It can be selected, for example, when clauses related to important e-services can
not be fulfilled and the process can not proceed.
In the right side of Figure 3, the Business Value List is composed as follows (Andrieux et al. 2007):
Penalty: defines an expression to be assumed when an associated objective is not met;

Figure 2. Feature Metamodel for e-Services, QoS and Control Operations
ISBN: 978-972-8939-25-0 2010 IADIS
Value Unit: (type xs:string) defines the unit for assessing penalty. It is used, in this work, to
represent a currency in case of penalty application or a control operation such as renegotiation or process;
Value Expr: (type xs:any) represents a value in case of penalty application or a control activity such
as variable, price or clause renegotiation;
Assesment Interval: defines the interval over which a penalty is assessed. It can be one of the
following: i) Time Interval that defines the assessment interval as a duration. For example, a weekly or
monthly interval for defining the assessment; and, ii) Count that defines the assessment interval as a service
specific count, such as the number of invocations.
When the Contract Renegotiation operation is chosen, a negotiation protocol must be specified. It will be
performed after a notification sent by the monitor to the collaborating parties. Other operations such as
Process Terminate, Process Rollback and Process Suspend will be executed by the WS-BPEL server. The
monitor and WS-BPEL server are elements of the infrastructure described in Section 5.
The proposed BPM infrastructure comprises four organizations: consumer, provider, negotiator and monitor.
The Consumer Organization includes: i) a structure called WS-Contract Definition responsible for
negotiation and establishment of WS-Contracts based on features; ii) a structure WS-Contract Execution
responsible for the business process execution; and, iii) a SOC System necessary if the consumer services
are part of the business process to be executed. In the Provider Organization, the SOC System control the
Web Services subcontracted by the consumer. The Monitor Organization has one structure WS-Contract
Monitoring that follows the business process execution guided by the QoS terms contained in the WS-
contract for service monitoring. The Negotiator Organization has one structure WS-Contract Negotiation
that uses a set of protocols responsible to (re)negotiation of contracts between providers and consumers.
The FeatureContract toolkit, extended in this work, implements the structure WS-Contract Definition and
has the following functionalities (Fantinato et al. 2008): i) Feature models creation and management; ii)
Services section, business process and agreement terms creation from the feature model; iii) Contract model
management through edition of its services section, business process and agreement terms; iv) Feature model
configuration that represents the services and related QoS levels to be contracted; and, v) e-contract
instantiation. The FeatureContract has a set of components that supports the different stages of contract
establishment, some were reused from other research and development groups whereas others were
developed by our research group. A description of each component is presented as follows:
FeaturePlugin: supports the elaboration of feature models and also the configuration of such
models. It was developed by Czarnecki research group (Antkiewicz & Czarnecki 2004), as an Eclipse plugin,
and it implements cardinality-based feature modeling. Only a few adaptations were necessary in order to
incorporate it as a component of the FeatureContract toolkit;

Figure 3. Mapping between control operation elements regarding both metamodels
IADIS International Conference WWW/Internet 2010
Feature Model Parser (FMParser): supports the automatic transformation from e-service feature
models to an initial version of a WS-contract template. It was specifically developed to be a component of
the FeatureContract toolkit. This component works as a document parser, it has been implemented as J ava
code that is responsible to transform the XML exported by the FeaturePlugin in the services section, business
terms and business process;
XML Editor: supports the graphical visualization and edition of XML specifications. Since there is
no particular tool for editing WS-Agreement specifications yet, this tool is used to edit the business terms
section, represented by the WS-Agreement definitions, of the WS-contract template, if necessary. It is an
Eclipse plugin, part of the Web Tools Platform (WTP) (Eclipse Foundation 2010b);
WSDL Editor: supports the graphical visualization and edition of WSDL specifications. It is used
to edit the services section, represented by the WSDL definitions, of the WS-contract template generated by
the FMParser, if necessary. It is also an Eclipse WTP plugin (Eclipse Foundation 2010b);
Eclipse BPEL: supports the graphical visualization and edition of WS-BPEL specifications (Eclipse
Foundation 2010a). It is used to edit the business process of the WS-contract template generated by the
FMParser, which is created in a basic way. It is always necessary to edit it to fix the process logic;
WS-Contract Factory: supports the automatic WS-contract instantiation, based on the WS-contract
template and on the e-services feature model configurations. It was specifically developed to be a component
of the FeatureContract toolkit. It works as a document parser, which removes all the parts of the template
linked with features not present in the feature models configurations. This removal process is carried out
following the hierarchical levels of the contract parts, i.e. when an upper-level template element has to be
removed, all its internal lower-level element are removed as well.
The FeatureContract toolkit is considered a prototype as not all the functionalities have been implemented
yet, but it can express the feasibility of this approach as showed in the next Section. The extensions proposed
here were developed and an application example was chosen in order to be a complex task, but due to lack of
space it was simplified.
This section presents a summary of an application example to show the feasibility of the proposed approach.
This example requires a contract to be established between two fictitious organizations: A Travel Agency and
an Airline Company. The Travel Agency (Consumer)'s system supports tourism packages sale by
composing third-party services; e-services provided by the travel agency include: customer notification,
offers advertisement and travel packages that consume Airline Company e-services. The Airline Company
(Provider)'s system supports flight ticket sale and booking; e-services provided by the airline company
include: flight booking and purchase; seat and food selection. The stages for e-contract establishment are
presented, only in the Airline Company perspective, as follows.
In the first stage (Services Feature Model Elaboration), each company must elaborate a feature model
to represent: the provided e-services; the QoS levels related to each service; and the control operations related
both to e-services and/or QoS levels. Figure 4(a) presents a feature model example for an Airline Company,
which has two service groups: flight services group and services selection group. This first group has the
following services: timetable query, flight purchase and flight booking; and the services selection group has
seat selection and food selection. Only some items are expanded. The Airline Company feature model gives
the consumer (Travel Agency) a structured overview of which e-services, QoS levels and control operations
are offered by the provider. During the feature model configuration stage, described in the following, with
the control operations associated to the feature model, the consumer can choose e-services and their QoS
levels with control operations to be applied when QoS levels are not met. With the services features models
elaborated, XML files can be exported by the FeaturePlugin to be used in e-contract template creation.
In the second stage (e-contract Template Creation), the XML files exported from the feature models are
used by the FeatureContract to extract the WSDL and WS-Agreement sections of the template. Moreover, the
Eclipse BPEL plugin is used to define the business process flow. However, BPEL and WSDL sections of the
e-contract are not addressed in this paper since the extension proposed here is more closely related with the
WS-Agreement section. The template is composed by all the services and related QoS levels offered to be
contracted with the respective control operations in case of contract violation, as present in Figure 4(a). The
ISBN: 978-972-8939-25-0 2010 IADIS
main idea is to create one template that can be instantiated many times with different feature configurations.
Figure 5 shows a WS-Agreement excerpt that represents renegotiation related to a QoS Level. Not all the
possible values are depicted due to lack of space. Latter they will be removed, only the level chosen in the
feature model configuration will be left as described in the following subsection.

Figure 4. (a) Feature model; and, (b) Feature model configuration for the airline company
In the third stage (Services Feature Model Configuration), the Travel Agency system selects the
e-services to be contracted from the Airline Company system, QoS levels and control operations related to
these services are also configured. As depicted in Figure 4(a) by the symbol , all the features of the flight
services group are mandatory, the symbols and represents optional features that can be chosen by the
contracting party. Figure 4(b) depicts an example of a feature model configuration. The control-operations
sub-tree is associated to the flight-purchase e-service and its QoS attributes. The renegotiation of price is
defined as the default control operation for the flight-purchase service, which must be triggered if any QoS
attribute, for which there is no specialized control operation, is not met. Latter in the e-contract instantiation
the penalty tags, related to the flight-purchase e-service, that represents renegotiation of clause and variable
will be removed from the e-contract template. Only the penalty that represents price renegotiation will be left.

Figure 5. WS-Agreement Template Excerpt
In the last stage (E-contract Instantiation), the feature models configurations of the Airline Company
and Travel Agency systems indicate which information are relevant to a specific e-contract instance. From
the feature models configurations, new XML files are generated. Using the WS-Contract Factory, of the
FeatureContract toolkit, all the information related to e-services, QoS attributes and control operations not
selected during feature models configuration are removed from the e-contract template. New WDSL,
WS-Agreement and WS-BPEL files, containing only the information necessary to the specific agreement,
form a new e-contract instance, including control operations information related to Web Services and QoS

IADIS International Conference WWW/Internet 2010
This paper has presented an infrastructure that supports e-contract negotiation, establishment, enactment and
renegotiation. The emphasis was in the FeatureContract toolkit and the control operations in the contract
metamodel. The main advantages of the approach are: i) efficient information management, organization and
reuse, necessary for the negotiation and renegotiation of e-contracts through feature modeling utilization; ii)
understanding of the renegotiation process through identifying all possible alternatives to dynamically adjust
the e-contract; iii) information organization and presentation of e-services and QoS attributes linked with
control operations using feature modeling; and, iv) extension of the WS-Contract metamodel to include the
control operations elements already supported by the feature models.
As some lessons learned, the use of feature models to represent control operations to handle e-contract
violations makes this process easier to understand, simple and systematic. It can bring benefits for distinct
stakeholders, at different levels, as it has a high level of abstraction and structured way for representing
information. However, some disadvantages or limitations of the approach can be pointed out: i) necessity of
knowledge about the feature modeling; and, ii) negotiation is still made in an offline way; negotiation
protocols have not yet been included to automatically perform negotiation.
Future works, besides focusing on the weaknesses cited above, include: full implementation of the
WS-Contract Negotiation element with some example protocols; and integration with the WS-Contract
monitoring tool which has been developed by the same research group.
This work was supported by The State of So Paulo Research Foundation (FAPESP) and The National
Council for Scientific and Technological Development (CNPq), Brazil.
Alves, A et al. 2007, Business process execution language for web services version 2.0, OASIS, Available from:
<>. [22 J uly 2010].
Andrieux, A et al. 2007, Web services agreement specification (ws-agreement), Open Grid Forum, Available from:
<>. [22 J uly 2010].
Angelov, S & Grefen, P 2008, An e-contracting reference architecture, J. Syst. Softw., Vol. 81, No. 11, pp. 18161844.
Angelov, S & Grefen, P 2008, Supporting the diversity of b2b e-contracting processes, Int. J. Electron. Commerce, Vol.
12, No. 4, pp. 3970.
Antkiewicz, M & Czarnecki, K 2004, FeaturePlugin: feature modeling plug-in for Eclipse, in Proc. of the 2004 OOPSLA
workshop on eclipse technology eXchange, New York, USA, pp. 6772.
Bacarin, E et al. 2008, Contract e-negotiation in agricultural supply chains, Int. J. Electron. Commerce, Vol. 12, No. 4, pp.
Czarnecki, K et al. 2005, Staged configuration through specialization and multi-level configuration of feature models, in
Software Process Improvement and Practice, Vol. 10, No. 2, pp. 143-169.
Eclipse Foundation 2010a, BPEL Project, Available from: <>. [22 J uly 2010].
Eclipse Foundation 2010b, Web Tools Platform, Available from: <>. [22 J uly 2010].
Fantinato, M et al. 2008, Ws-contract establishment with qos: an approach based on feature modeling, Int. J. Cooperative
Inf. Syst., Vol. 17, No. 3, pp. 373407.
Fantinato, M et al. 2010, Product line in the business process management domain, in Applied Software Product Line
Engineering, eds K C Kang, V Sugumaran, & S Park, Auerbach Publications, pp. 497530.
Grefen, P et al. 2001, CrossFlow: Cross-Organizational Workflow Management for Service Outsourcing in Dynamic
Virtual Enterprises, Data Engineering Bulletin, pp. 5257.
Papazoglou, M 2007, Web Services: Principles and Technology, Prentice Hall.
Pohl, K et al. 2005, Software Product Line Engineering: Foundations, Principles and Techniques, 1st ed, Springer.
ISBN: 978-972-8939-25-0 2010 IADIS
Niklas Eriksson* and Peter Strandvik**
*Arcada University of Applied Sciences, Jan-Magnus Janssonin aukio 1, 00550 Helsinki Finland
**bo Akademi University, Institute for Advanced Management Systems Research, Jouhkahainenkatu 3-5A Turku,
This paper presents results from a study of the intended use of mobile tourism services. The theoretical foundation was
drawn from the unified theory for the acceptance and use of technology (UTAUT) and from relevant research theories for
adoption of electronic and mobile services. The data was collected through a self-administered questionnaire which was
handed out to 103 university students. The findings suggest that value expectancy has the most significant direct impact
on the intended use of mobile tourism services. Moreover, attitudes towards mobile technology and social influence have
a significant direct impact on the intended use of mobile tourism services.
Mobile services, tourism, technology adoption.
The use of the Internet for doing commerce or interacting with customers has been growing rapidly in the
world wide tourism industry. Mobile commerce, or e-commerce over mobile devices, on the other hand has
had many conflicting predictions on its future popularity. Most predictions have been overly optimistic.
However, the benefits that arise from mobile technology have not yet been fully delivered, which to some
extent is explained by the fact that mobile applications, due to complexity or lack of relevance, fail to meet
customers expectations (Carlsson et al. 2006). Travel and tourism is an industry in which several different
projects have been conducted where mobile applications have been developed, tested and implemented, some
even with moderate success (e.g. Schmidt-Belz et al 2003, Repo et al 2006, Riebeck et al. 2009). Moreover,
there are a few different studies (e.g. Hpken et al 2007, Riebeck et al 2009, Katsura & Sheldon 2009) made
on consumer adoption of mobile tourism services. Nevertheless, more is needed in order to fully understand
the adoption issues of mobile services in different contexts.
The New Interactive Media (NIM) project, with funding from the European Union and the regional
government of the land islands, was a development programme of increasing knowledge, production and
use of new interactive media on the land Islands
in Finland during the years 2006 2008. Within the
project several mobile applications were developed for the travel and tourism sector on the islands. Two of
them, MobiPortal and MobiTour, will in this study be used as example services to investigate determinants
with direct effect the intended use of mobile tourism services.
The services in the NIM project were planned with a common logic namely the Braudel rule: freedom
becomes value by expanding the limits of the possible in the structures of everyday life (as presented by
Keen & Mackintosh 2001). The rule was then translated into a tourism setting which means that tourists real

land is an autonomous and unilingual Swedish region in Finland with its own flag and approximately 26.700 inhabitants. land is
situated between Finland and Sweden and consists of 6 500 islands. (
IADIS International Conference WWW/Internet 2010
or perceived need had to be met by the services and moreover, the services need to profoundly change the
way a tourist does or experience something
MobiPortal is a mobile version of an informa
site of the land Islands. The portal includes search for events, restaurants etc., a map service and facts on
the land Islands.
MobiTour is a guide for attractions such as the Bomarsund fortress which is downloadable / streamable to
the visitors own devices. The guide includes voice and/or video guidance.
See user interface of the two services in figure 1.

Figure 1. User interface of MobiPortal and MobiTour
Both services ought to expand the limits of a tourist according to the Braudel rule by enabling 1) instant
access to local information and 2) experience enhancement for certain unmanned attractions.
experience enhancement features are generally seen as key drivers for successful customer satisfaction in
tourism (Pine & Gilmore 1999). The determinants for the intended use of mobile tourism services are,
however, a complex issue which will b
Several models of technology adoption have been developed. One of the most used models is the technology
acceptance model (TAM) by Davis (1989) which is based on the theory of reason action (TRA) by Fishbein
et al. (1975). Other often used models in techno
theories (DIT) by Rogers (1995) and the unified theory for the acceptance and use of technology (UTAUT)
by Venkatech et al. (2003) which combines TAM with several other adoption models e.g. DIT. The U
model states that there are three determinants performance expectancy, Effort expectancy and Social
Influence with direct effect on the behavioral intention of technology. Two determinants facilitating
conditions and behavioral intentions have a direc
determinants self-efficacy, attitudes towards technology and anxiety are not direct determinants of the
intended behavior. Moreover, four moderators gender, age, experience and voluntariness of use are inc
in the UTAUT model. See figure 2
direct effect on intentional behavior: performance expectancy (PE), Effort expectancy (EE) and Social
Influence (SI), as our aim is primarily to f
intention of mobile tourism services. Nevertheless, due to that UTAUT is primarily focusing on
organizational contexts and not on consumer contexts; we will modify PE, EE and SI and include the
determinants Anxiety (AX) and Attitude (ATT) in our test. Moreover, based on previous tests of UTAUT a
modification to the model should be made when investigating mobile services (Carlsson et al 2006). Hereby
we discuss relevant research theories for adopti
modification of UTAUT for PE, EE, SI, AX and ATT. The moderators are left out in this research paper.

to be met by the services and moreover, the services need to profoundly change the
way a tourist does or experience something and to the better (Harkke 2007).
is a mobile version of an information portal which is the official tourist
site of the land Islands. The portal includes search for events, restaurants etc., a map service and facts on
is a guide for attractions such as the Bomarsund fortress which is downloadable / streamable to
the visitors own devices. The guide includes voice and/or video guidance.
See user interface of the two services in figure 1.
MobiPortal MobiTour

Figure 1. User interface of MobiPortal and MobiTour
Both services ought to expand the limits of a tourist according to the Braudel rule by enabling 1) instant
access to local information and 2) experience enhancement for certain unmanned attractions.
experience enhancement features are generally seen as key drivers for successful customer satisfaction in
tourism (Pine & Gilmore 1999). The determinants for the intended use of mobile tourism services are,
however, a complex issue which will be discussed next.
Several models of technology adoption have been developed. One of the most used models is the technology
acceptance model (TAM) by Davis (1989) which is based on the theory of reason action (TRA) by Fishbein
et al. (1975). Other often used models in technology adoption research are the diffusion of innovations
theories (DIT) by Rogers (1995) and the unified theory for the acceptance and use of technology (UTAUT)
by Venkatech et al. (2003) which combines TAM with several other adoption models e.g. DIT. The U
model states that there are three determinants performance expectancy, Effort expectancy and Social
Influence with direct effect on the behavioral intention of technology. Two determinants facilitating
conditions and behavioral intentions have a direct effect on the actual use of technology and three
efficacy, attitudes towards technology and anxiety are not direct determinants of the
intended behavior. Moreover, four moderators gender, age, experience and voluntariness of use are inc
See figure 2. The focus in this study will be to examine determinants in UTAUT with
direct effect on intentional behavior: performance expectancy (PE), Effort expectancy (EE) and Social
Influence (SI), as our aim is primarily to find out determinants with direct influence on the behavioral
intention of mobile tourism services. Nevertheless, due to that UTAUT is primarily focusing on
organizational contexts and not on consumer contexts; we will modify PE, EE and SI and include the
eterminants Anxiety (AX) and Attitude (ATT) in our test. Moreover, based on previous tests of UTAUT a
modification to the model should be made when investigating mobile services (Carlsson et al 2006). Hereby
we discuss relevant research theories for adoption of electronic and mobile services to motivate the
modification of UTAUT for PE, EE, SI, AX and ATT. The moderators are left out in this research paper.
to be met by the services and moreover, the services need to profoundly change the
which is the official tourist
site of the land Islands. The portal includes search for events, restaurants etc., a map service and facts on
is a guide for attractions such as the Bomarsund fortress which is downloadable / streamable to

Both services ought to expand the limits of a tourist according to the Braudel rule by enabling 1) instant
access to local information and 2) experience enhancement for certain unmanned attractions. Especially
experience enhancement features are generally seen as key drivers for successful customer satisfaction in
tourism (Pine & Gilmore 1999). The determinants for the intended use of mobile tourism services are,
Several models of technology adoption have been developed. One of the most used models is the technology
acceptance model (TAM) by Davis (1989) which is based on the theory of reason action (TRA) by Fishbein
logy adoption research are the diffusion of innovations
theories (DIT) by Rogers (1995) and the unified theory for the acceptance and use of technology (UTAUT)
by Venkatech et al. (2003) which combines TAM with several other adoption models e.g. DIT. The UTAUT
model states that there are three determinants performance expectancy, Effort expectancy and Social
Influence with direct effect on the behavioral intention of technology. Two determinants facilitating
t effect on the actual use of technology and three
efficacy, attitudes towards technology and anxiety are not direct determinants of the
intended behavior. Moreover, four moderators gender, age, experience and voluntariness of use are included
. The focus in this study will be to examine determinants in UTAUT with
direct effect on intentional behavior: performance expectancy (PE), Effort expectancy (EE) and Social
ind out determinants with direct influence on the behavioral
intention of mobile tourism services. Nevertheless, due to that UTAUT is primarily focusing on
organizational contexts and not on consumer contexts; we will modify PE, EE and SI and include the
eterminants Anxiety (AX) and Attitude (ATT) in our test. Moreover, based on previous tests of UTAUT a
modification to the model should be made when investigating mobile services (Carlsson et al 2006). Hereby
on of electronic and mobile services to motivate the
modification of UTAUT for PE, EE, SI, AX and ATT. The moderators are left out in this research paper.
ISBN: 978-972-8939-25-0 2010 IADIS

Figure 2. The final UTAUT model
3.1 Performance Expectancy (Value Expectancy)
The TAM model proposes two determinants, perceived usefulness and perceived ease of use, which impact
the intended usage behavior of a system and the adoption behavior as a result (Davis 1989). The first one,
perceived usefulness, is defined as the degree to which a person believes that using a particular system
would enhance his or her performance. In the UTAUT model perceived usefulness has emerged into
performance expectancy and defined as the degree to which an individual believes that using the system will
help him or her to attain gains in job performance. However, perceived usefulness and performance
expectancy are foremost designed to research work performance improvements in organizational contexts. In
consumer markets consumer behavior is also influenced by other factors. It is typical that non-efficiency
factors impact consumer adoption of technology, e.g. good tourist technologies are not only those that make
tourists more efficient, but that also make tourism more enjoyable. Thus tourism can be characterized as
wandering, where tourists attempt to enjoy the city environment and chance upon things of interest, rather
than optimizing (Brown & Chalmers 2003). As the mobility (on the move) capability is generally seen as the
key value driver in m-commerce (Anckar & Eriksson 2003), mobile technology clearly has the potential to
support the wandering aspect of tourism. A word like flexibility has commonly been used to describe the
independence of time and space that is provided by mobile technology. According to Kim et al. (2005) the
hedonic motivation or the enjoyment aspect of tourism has, however, not been clearly defined in mobile
technology acceptance models. The perceived type and degree of perceived value of a mobile service depend
on the other hand on the situation or context of usage (Mallat et al 2006, Lee & Jun, 2005). Anckar & Dincau
(2002) introduced an analytical framework that identifies the potential value creating features of mobile
commerce. Mobile value elements in the framework for consumers on the move are: Time-critical
arrangements, Spontaneous needs, Entertainment needs, Efficiency ambitions and Mobile situations. Time-
critical arrangements refer to applications for situations where immediacy is desirable (arise from external
events), e.g. receive alerts of a changed transport schedule while on tour. Spontaneous needs are internally
awakened and not a result of external events, e.g. find a suitable restaurant while wandering around.
Entertainment needs, killing time/having fun, especially in situations when not being able to access wired
entertainment appliances, e.g. kill or fill time in transportation. Efficiency ambitions aim at productivity, e.g.
use dead spots during a travel to optimize time usage. Mobile situations refer to applications that in essence
are of value only through a mobile medium (e.g. localization services), which ought to be the core of mobile
commerce. Consequently perceived mobile value represent the degree to which a person perceives value
arising from the mobility of the mobile medium (when not being able to access a stationary PC).
Nevertheless not only the mobile medium creates value for the consumer but the essence of the services
as well. Electronic information systems can be divided into productivity oriented and pleasure (hedonic)
oriented systems (van der Heijden 2004). The hedonic systems focus on the fun-aspect of using an
information system rather than on productive use. For example, as discussed, good tourist technologies are
not only those that make tourists more efficient, but that also make tourism more enjoyable
As a result of the above discussion we will refer to performance expectancy as value expectancy which
includes mobile value and service value as productivity value and pleasure value. We expect value
expectancy to have a direct effect on intentional behavior.

Effort expectancy

Social Influence
Behavioral Intentions Use Behavior



Voluntariness of Use

IADIS International Conference WWW/Internet 2010
H1: Value expectancy has a significant direct effect on intentional behavior of mobile tourism services
3.2 Effort Expectancy
The second TAM determinant perceived ease of use is defined as the degree to which a person believes that
using a particular system would be free of effort. In the UTAUT model the ease of use determinant has
emerged into effort expectancy and defined as the degree of ease associated with the use of the system.
Moreover, the ease of use aspect has been widely discussed in mobile commerce. Limitations of mobile
devices (e.g. screen size) cause consumers to hesitate whether to adopt mobile commerce or not. According
to Cho et al (2007) device limitations suggest that focusing on easy to use mobile applications could enhance
the consumer acceptance of mobile commerce. Kaasinen (2005) points out that mobile services need to be
easy to take into use as well (ease of adoption) as mobile services are typically used occasionally and some
services may be available only locally in certain usage environments. As a consequence, information on
available services should be easy to get and the services should be easy to install and to start using. The ease
of adoption aspect has not been taken into account in models such as TAM. The ease of taking a service into
use may in fact have a direct impact on the adoption behavior of a mobile service (Kaasinen 2005). On the
other hand when problems arise, users in the consumer market are often expected to solve the problems on
their own (Repo et. al 2006). Consequently the use may rely on proper instructions or on a helping hand from
someone. Proper support conditions also in a consumer market may therefore be important especially for
advanced mobile services. This aspect is in the UTAUT called facilitating conditions which is proven to
significantly influence the actual use of technology in organizational contexts. Nevertheless we argue that
consumers many times expect to take a new product or service into use without instructions or help,
especially when it will be only temporary used while visiting a tourist destination. Furthermore, facilitating
conditions didnt have a direct link to the actual use of mobile services in a previous study by Carlsson et al
(2006). Therefore ease of adoption and facilitating condition will here together with ease of use be included
in the effort expectancy determinant. Hereby we expect effort expectancy to have a direct effect on
intentional behavior.
H2: Effort expectancy has a significant direct effect on intentional behavior of mobile tourism services
3.3 Social Influence
In UTAUT social influence has been identified to influence the intended use of information technology. In
UTAUT social Influence is defined as the degree to which an individual perceives that important others
believe he should use the new system (Venkatech et al., 2003). Social influence is also known as subjective
norm in the theory of reason action (Fishbein et al 1975) and in its extension theory of planned behavior
(Arjzen 1991). In consumer markets image and social status have been proposed to impact consumers
intentions to adoption of a WAP-enabled mobile phone (Teo & Pok 2003). A positive or negative subjective
norm can in fact in consumer markets make or break a new technology based product (Schepers and Wetzels
2007). Hereby we expect social influence as image to have a direct effect on intentional behavior.
H3: Social Influence has a significant direct effect on intentional behavior of mobile tourism services
3.4 Anxiety
Anxiety in UTAUT has been defined as evoking anxious or emotional reactions when it comes to performing
a behaviour (e.g. using a computer), but has not been seen to have a direct influence on intentional behaviour
of information technology. Therefore anxiety has not been included in the final UTAUT model. According to
Rogers (1995),The innovation-decision is made through a cost benefit analysis where the major obstacle is
uncertainty. Perceived risk is commonly thought of as felt uncertainty regarding possible negative
consequences of using a product or service and has been added to the two TAM determinants as a negative
influencer on intended adoption behaviour (Featherman & Pavlou 2003). Trust, as trust in the service vendor
to minimize the risks, has also been added to the TAM model (e.g. Cho et al 2007, Kaasinen 2005) and
pointed out as a strong influence on the intended use of mobile services due to that mobile commerce is still
at its initial stage (Cho et al. 2007). We refer to anxiety and trust as perceived risk defined by Featherman &
Pavlou 2003. They divide the perceived risk for electronic services into the following elements; performance
ISBN: 978-972-8939-25-0 2010 IADIS
risk, financial risk, time risk, psychological risk, social risk and privacy risk. Performance risk refers to the
possibility of a service to malfunction and not performing as it was designed and advertised. The financial
risk refers to the potential monetary outlay associated with the initial purchase price as well as the subsequent
maintenance cost of the product and the possibility of fraud. Time risk refers to that the consumer may lose
time when making a bad purchasing decision e.g. by learning how to use a product or service only to have to
replace it if it does not perform to expectations. Psychological risk refers to the potential loss of self-esteem
(ego loss) from the frustration of not achieving a buying goal. Social risk refers to potential loss of status in
ones social group as a result of adopting a product or service, looking foolish or untrendy. Privacy risk refers
to the potential loss of control over personal information, such as when information about you is used without
your knowledge or permission. At least security and privacy issues have been highlighted as barriers to
mobile commerce (ODonnell et al. 2007). Also financial risks in form of high costs, including operating
costs and initial costs, have been highly ranked by consumers as hindrances for m-commerce in its early
stages (Anckar et al. 2003). Hereby we expect anxiety (perceived risk) to have a direct effect on intentional
behavior. The element of social risk is already included in the determinant social influence as described
earlier and is therefore not included in Anxiety.
H4: Anxiety has a significant direct effect on intentional behavior of mobile tourism services
3.5 Attitude towards using Mobile Technology
In UTAUT Attitudes towards technology has been defines as an individual's overall affective reaction to
using a system. Attitude has been theorized as having no direct effect on intentional behavior of information
technology. Also in diffusion theories personal innovativeness has been argued to be a prior condition to
other determinants of adoption decisions (Rogers 1995). Mobile service studies, nevertheless, argue that
attitude of mobile technology in fact may have a direct effect on the intended behavior of mobile services
(Carlsson et al 2006). Herby we expect attitudes towards using mobile technology to have a direct effect on
intentional behavior. We see three relevant elements of the attitude aspects: try out of mobile technology in
general, willingness of purchasing the latest mobile devices and curiosity towards new mobile services.
H5: Attitudes towards using mobile technology has a significant direct effect on intentional behavior of
mobile tourism services
3.6 Behavioral Intention
Information technology acceptance models separate between intended use and actual use. As we are only
demonstrating examples of mobile tourism services to the respondents, without them being able to use the
services on their own, we cannot expect the respondents to give answers based on their actual use but only on
their intended use of similar services. Therefore we need to investigate possible determinants which effect
(positively or negatively) the intended use of mobile tourism services. Nevertheless the intended use in
acceptance models has a direct effect on the actual use. We see three aspects of the intended use: the intended
use in the near future (next time visiting e.g. land islands), believe in future use when visiting different
places and the intended use compared to use of traditional services (e.g. paper brochures and guides).
Based on the literature discussion possible determinants for the intended consumer adoption of mobile
tourism services are: value expectancy, effort expectancy, social influence, anxiety and attitude towards
mobile technology. The model to be tested is shown in figure 3.

Figure 3. Research model
Value expectancy
Effort expectancy

Intentions to use mobile
tourism services
Social Influence


H5 Attitude towards
mobile technology

IADIS International Conference WWW/Internet 2010
Based on the literature discussion item statements were developed for each determinant. In total 19 item
statements were measured on a five point Likert scale (1 = I totally disagree and 5 = I totally agree). Negative
statements were reversed in the analysis so that all scales were comparable. The used statements are viewed
in Table 1.
Data was collected at Arcada University of Applied Sciences in Helsinki Finland during May 2010. The
questionnaire was handed out in a class situation and included a brief introduction to the two example mobile
tourism services; MobiTour and MobiPortal. The sample included in total 103 students, with a female student
domination of 62,1% . The average age was 21,5 years and 38,8% use mobile Internet services at least a few
times a month.
We used the software Smart PLS 2.0 to analyze the data. Partial Least Square (PLS) path modeling is
very well suited for analyzing small samples (Chin & Newstedt 1999). PLS is also increasingly being used
within several research fields such as research of consumer behavior (Hensler et al. 2009). Thereby PLS is
well motivated to use for this study. A two-stage analytical procedure was conducted to assess the
measurement model and then the structural relationships.
Table 1. Statements of the determinants
Determinants Items Item statements, scale 1 5
Value expectancy

Mobile value

I would find services like MobiPortal / MobiTour useful when on the move (when I
cant access a stationary PC)
Productivity value Using services like MobiPortal / MobiTour would save me time
Pleasure value Using services like MobiPortal / MobiTour would be fun
Effort expectancy

Ease of use It would be easy for me to become skillful at using services like MobiPortal /
Ease of adoption

In my opinion services like MobiPortal / MobiTour would be easy to take into use
(access / install and learn to use)
Facilitating condition I wouldnt need anyone to help me use services like MobiPortal / MobiTour
Social Influence Image / social status I feel trendy if I use services like MobiPortal / MobiTour
*Reverse scale

Performance risk I dont trust services like MobiPortal / MobiTour to perform as it is designed and
Financial risk 1

I fear that the monetary layout (e.g. transaction costs) would be high for services like
MobiPortal / MobiTour
Financial risk 2 I fear that I would need to pay without being able to use the service anyway
Time risk I fear that services like MobiPortal / MobiTour would waste my time
Psychological risk I believe I would be frustrated if services like MobiPortal / MobiTour do not perform
as promised
Privacy risk Im afraid of losing control over personal information when using services like
MobiPortal / MobiTour
Attitudes towards
Mobile device I want my mobile phone to be the latest model
Mobile technology I like to try out the latest mobile technologies
Mobile service Im curious about using new mobile phone services

Near future I intend to use services like MobiPortal / MobiTour next time I visit e.g. the land
Traditional use I would rather use services like MobiPortal / MobiTour than traditional ones (e.g.
paper brochures and guides)
Future use I believe I will use services like MobiPortal / MobiTour in the future when I visit
different places
The results are assessed by analyzing the correlations between measures of potentially overlapping
determinants. Therefore we calculated the average variance extracted (AVE) and the composite reliability to
assess determinant reliability. AVE shared between each determinant and its measures should be greater than
the variance shared with other determinants. All determinants expect for Anxiety showed good composite
reliability (>0.7) and AVE (>0.5). The specific items should load heavier on their own determinants than the
other determinants. All items of the model showed the highest loadings for their own determinant. The
ISBN: 978-972-8939-25-0 2010 IADIS
loading per item for a determinant should also exceed 0,55 and their significance should exceed 1,96. That
was not the case for a few items. Within the anxiety determinant the financial risk 1 and 2 items plus
psychological risk and privacy risk showed very low path loadings and low significance. Moreover both
AVE and composite reliability were low for anxiety and therefore these items were excluded from the final
The latent scores were then calculated. Value expectancy and Attitude towards using mobile technology
show the highest significant impact on the behavioral intention. Especially the determinant value expectancy
stands out with a loading of 0,306. Also social influence has a significant impact on the intentional behavior.
Therefore the hypothesis H1, H3 and H5 are supported in this study. Nevertheless and surprisingly H2 and
H4 are rejected as not significant. The R2 value for Intentional behavior is 0,522 which means that about
52,2% of the variance in intentions to use mobile tourism services can be explained by Value expectancy,
Social influence and Attitude. The R2 value is good compared to e.g. earlier TAM studies. See figure 3.

*** P <0,001, **P < 0,01,*P < 0,05, n.s. = not significant
Figure 3. Tested research model
It must be emphasized that the sample consisted of students who may differ in their behavior compared to a
wider population. Students may be more used to technology and their effort expectancy and anxiety towards
technology services may therefore be different than for other user segments. Therefore the rejection of H2
and H4 shouldnt be over generalized for a wider population. On the other hand it can be argued that mobile
commerce is not in its initial stage anymore, which has been a key argument at least for consumer anxiety.
Value expectancy showed the highest direct impact on the intention to use mobile tourism services. Many
earlier studies have emphasized, as discussed in the theoretical foundation, the importance of value when
marketing and designing mobile services. This study supports that the value expectancy determinant is the
key driver also for the intended use of mobile tourism services, at least when it comes to young university
adults in Finland. This study also indicates that attitudes towards mobile technology and social influence play
a significant role in the adoption process of mobile tourism services.
Future research should include a wider sample of respondents. Moderators such as gender, age and usage
experience should be taken into account as well in the analysis. Therefore we plan to extend the sample to
e.g. incoming tourists on the land islands.
Anckar, B. and DIncau, D., 2002. Value creation in mobile commerce: Findings from a consumer survey. Journal of
Information Technology Theory and Application, Vol. 4, pp. 43-65.
Anckar, B., Eriksson, N. 2003. Mobility: The Basis for Value Creation in Mobile Commerce?. Proceedings of the
International Conference SSGRR03. Telecom Italia learning Services, LAquila, Italy.
Value expectancy

Effort expectancy

Intentions to use mobile
tourism services
R2 = 0,522
Social Influence


Attitude towards
mobile technology

0,115 n.s.
0,138 n.s.
IADIS International Conference WWW/Internet 2010
Anckar, B., Carlsson C., Walden, P. 2003. Factors affecting consumer adoption decisions and intents in mobile
commerce: empirical insights. Proceedings of the 16
Bled eCommerce conference. Bled, Slovenia.
Ajzen, I. 1991. The Theory of Planned Behavior. Journal of Organizational Behavior and Human Decision Processes.
No. 50, pp. 179-211.
Brown, B. and Chalmers, M., 2003. Tourism and mobile technology. Proceedings of ECSCW 2003. Helsinki, Finland,
pp. 335-355.
Brown B. et al, 2005. Sharing the square: collaborative leisure in the city streets. Proceedings of ECSCW 2005. Austria.
Carlsson, C., Carlsson, J., Hyvnen, K., Puhakainen, J., Walden, P. 2006. Adoption of Mobile Devices/Services
Searching for Answers with the UTAUT. Proceedings of the 39th HICCS 2006. IEEE Computer Science Press.
Chin, W. W., & Newsted, P. R. 1999. Structural equation modeling analysis with small samples using partial least
squares. In: R. H. Hoyle (Ed.), Statistical strategies for small sample research. Thousand Oaks, CA: Sage.
Cho, D., Kwon, H., Lee, H-Y. 2007. Analysis of trust in Internet and mobile commerce adoption. Proceedings of the 40th
Hawaii International Conference on System Sciences. IEEE Computer Science Press.
Davis, F. D., 1989. Perceived Usefulness, Perceived Ease of Use and User Acceptance of Information Technology. MIS
Quarterly. Vol. 13, No.3, pp. 319-340.
Featherman, M., Pavlou, P. 2003. Predicting e-services adoption: a perceived risk facets perspective. Journal of Human
Computer studies, No. 59, pp. 451-474.
Fishbein, M., Ajzen, I., 1975. Belief, attitude, intention and behavior: an introduction to theory and research. Reading,
MA Addison-Wesley.
Harkke, V. 2007. NIM land: the quest for useful mobile services. Proceedings of the 20th Bled eConference. Slovenia.
Henseler, J., Ringle, C. M., Sinkovics, R. R. 2009. The use of partial least squares path modeling in international
marketing, in: Sinkovics, R. R. / Ghauri, P. N. (eds.), Advances in International Marketing (AIM), Vol. 20, Bingley,
pp. 277-320.
Hpken, W., Rasinger, J., Fuchs, M. 2007. Information Search with Mobile Tourist Guides: A Survey of Usage Intention.
Journal of Information Technology & Tourism. Vol. 9, No. 3-4, pp. 177-194(18).
Kaasinen, E., 2005. User acceptance of mobile services value, ease of use, trust and ease of adoption. Doctoral
Dissertation. 566 VTT, Espoo, Finland.
Keen, P., Macintosh, R. 2009. The Freedom Economy Gaining the M-commerce Edge in the Era of the Wireless
Internet. McGraw-Hill, Berkeley, USA.
Katsura, T., Sheldon, P. 2009. Forecasting mobile technology use in Japanese Tourism. Journal of Information
Technology & Tourism. Vol. 10, No 3.
Kim, D., Park, J., Morrison, A. 2005. A model of Tourist acceptance for mobile technology. Submitted to the 11th annual
conference of Asia Pacific Tourism Assoc.
Lee, T., Jun, J. 2005. Contextual perceived usefulness? Toward an understanding of mobile commerce acceptance.
Proceedings of the International Conference on Mobile Business. IEEE Computer Science Press.
Mallat, N., Rossi, M., Tuunainen, V., rni, A 2006. The impact of use situation and mobility on the acceptance of
mobile ticketing services. Proceedings of the 39th Hawaii International Conference on System Science.
O'Donnell, J., Jackson, M., Shelly, M., Ligertwood, J. 2007. Australian Case Studies in Mobile Commerce. Journal of
Theoretical and Applied Electronic Commerce Research. Vol. 2 No. 2, pp. 1-18.
Pine, J. and Gilmore, J., 1999. The Experience Economy. Pine and Gilmore Press.
Repo, P., Hyvnen, K., Saastamoinen, M. 2006. Traveling from B2B to B2C: Piloting a Moblog Service for Tourists.
Proceedings of the International Conference on Mobile Business. IEEE Computer Science Press.
Riebeck, M., Stark, A., Modsching, M., Kawalek, J. 2009. Studying the user acceptance of a Mobile Information System
for Tourists in the field. Journal of Information Technology & Tourism. Vol. 10, No 3.
Rogers, E., M 1995. Diffusion of Innovations. Free Press, 4th edition, New York, USA.
Schepers, J., Wetzels, M. 2007. A meta-analysis of the technology acceptance model: Investigating subjective norm and
moderation effects. Journal of Information Management. Vol. 44, No. 1, pp. 90 103.
Schmidt-Belz, B., Zipf, A., Poslad, S., Laamanen, H.2003. Location-based mobile tourist services first user
experiences: In: Frew, A. J., Hitz, M. & OConnor, P. (eds.) Information and Communication Technologies in
Tourism. Springer, Wien.
Teo, T.S.H., Pok, S.H. 2003. Adoption of WAP-Enabled Mobile Phones among Internet Users. Omega No. 31, pp. 483-
Van der Heijden, H. 2004. User Acceptance of hedonic information systems. MIS Quarterly Vol. 28, No. 4, pp. 695-704.
Venkatesh, V., Morris, M. G., Davis, G. B., Davis, F. D. 2003. User acceptance of information technology: toward a
unified view. MIS Quarterly. Vol. 27, No. 3, pp. 425 478.
ISBN: 978-972-8939-25-0 2010 IADIS
Cldio Martins, J oo Rosa, Laerte Franco and J orge Barbosa
PIPCA/Vale do Rio dos Sinos University - So Leopoldo RS Brazil
In mobile computing environments, the tracking of users allows applications to adapt to the contexts visited by users
(Context Awareness). In recent years, the use of context information and users profiles has been considered an
opportunity for ubiquitous commerce. The improvement and the wide adoption of location systems are stimulating the
tracking of users, allowing the use of Trails. A trail is the history of the contexts visited by a user during a period. This
article proposes a systemfor trails management and its application in the ubiquitous commerce. In this text, we consider
that Trail-Aware is an evolution of the simple use of contexts and profiles. The text presents a prototype and its
application in a shopping environment for discovery of deal opportunities using trail-aware.
Ubiquitous Computing, Ubiquitous Commerce, Location System, Trail, User Profile, Business.
Currently, the evolution of mobile devices and high-speed wireless networks has been stimulating researches
related to Mobile Computing (Diaz et al, 2010; Satyanarayanan et al, 2009). In this area, the improvement
and proliferation of Location Systems (Hightower and Borriello, 2001; Hightower et al, 2006) have
motivated the adoption of solutions that consider the users precise location in the providing of services
(Location-Based Services (Dey et al, 2010; Vaughan-Nichols, 2009)). However, lately, mobile applications
have also become to take into account the users current context to distribute content and services (Context
Awareness (Baldauf et al, 2007; Hoareau and Satoh, 2009)).
The state-of-the-art context-aware computing enables applications taking decisions based on users data
and their current contexts (Hoareau and Satoh, 2009). Nonetheless, we understand that would be valuable
also to consider the users past actions performed in the contexts visited during a period, such as the activities
did, the applications used, the contents accessed, and any other possible event. We believe that this would
improve the distribution of content and services in context-aware environments, because instead of using
only context and profile data, applications would be using the historic of the users actions to take decisions.
Studies on users monitoring in mobile computing systems have shown that is possible to record the
history of contexts visited by users and their actions performed in each context (Smith et al, 2006). Usually,
this history is called Trail (Driver and Clarke, 2008; Levene and Peterson, 2002). Thus, in context-aware
applications, trails would enable using the users past behavior to distribute content and services. Hence, we
propose the application of trails to context-aware computing, considering that Trail-Aware is an advance in
Context-Aware computing. In this article, we introduce a system, which is named TrailM, for trails
management in mobile computing environments.
An important application area of the joint use of context and profile is the commerce (Shekar et al, 2003;
Roussos and Moussouri, 2004). In this field, some applications distribute contextualized deal opportunities
(Franco et al, 2009). In the same way that occurs in context-aware applications, in the commerce domain,
applications are limited to take decisions based on profile and context information. Furthermore, the use of
dealers profiles in context-aware environments is a recent research topic (Franco et al, 2009). Nevertheless,
deal discovery guided by the dealers trails is still an opened research area. Hence, we propose the use of
trails in the commerce scenario, expecting to improve the deal opportunities. In this article, we present a case
IADIS International Conference WWW/Internet 2010
study in which TrailM was integrated with a ubiquitous commerce system (UbiTrade (Franco et al, 2009))
and used for contextualized discovery of deal opportunities.
The article is organized into six sections. Sect. 2 describes the TrailM architecture. The third Sect.
discusses the TrailM use in the ubiquitous commerce guided by trails. Sect. 4 presents the TrailM and the
UbiTrade prototypes as well as the integration between them, and still describes and evaluates the results
obtained from a case study involving trails in discovery of deal opportunities. Sect. 5 discusses related works
and the contribution of this work. And finally, Sect. 6 presents the conclusions and the future works.
This section presents the trails management system. Section 2.1 describes the main terms used by the system,
mainly those related to the trails composition, and finally, Section 2.2 presents the system architecture.
2.1 Entities, Trails, Ptrail, and Trailpoint
TrailM was designed to manage entities trails. In this context, entities represent persons, accessing
computational resources (for example, a smartphone), or mobile objects (for example, a vehicle). We choose
entity, instead of user, because then the model could manage trails of any mobile object. In this scenario,
different applications can be developed for trails. For example, a vehicle could have a trail, recording
information considered relevant, such as the occurrences of maintenances or failures.
Applications that use TrailM must temporally store data that composes a trail. For this reason, we propose
that the applications have internally a structure called ptrail (piece of trail), which is composed of the
following attributes: entity, resource, event, context, and location. The values of the entity and resource
attributes are stored statically in the ptrail after that an entity logs in the model. Nonetheless, the values of the
event, context, and location attributes are automatically updated as the actions or movements made by the
entity and monitored by the application. The records related to the contexts visited by an entity are stored in
only one trail. The trail is composed of a sequence of records of the ptrail. A record is composed of seven
attributes organized into three categories: identification, content, and temporality. The first category contains
only one attribute to identify the trail. The second category is used to store the ptrail content. The last
category contains date/time attribute, representing the time of the record creation.
We named trailpoint the process that makes the trail composition. This process sends the values
contained in the ptrail to be recorded in the entitys trail, which is stored in a server. The trailpoint occurs
when the application automatically identifies that an entity performed an event, for example, entry or exit
from a location, interaction with another entity, access to a file, among others. Figure 1 shows the trail
creation of a mobile device user. Each star represents the occurrence of an event, which causes a trailpoint.

Figure 1. Example of trail composition
2.2 The System Architecture
2.2.1 General Organization
Figure 2 presents a general view of the system, which is composed of a server (TrailM Server) and a client
(TrailM Client). In addition, TrailM considers the existence of an external provider that provides location
information. Next sections describe all these components, justifying why they are necessary.
ISBN: 978-972-8939-25-0 2010 IADIS

(a) TrailM Server (b) TrailM Client
Figure 2. The TrailM architecture
2.2.2 TrailM Server
Some of the TrailM Server basic tasks are to manage the entities trails and to enable application querying the
trails. To perform these functions, we designed the TrailServices module (see Figure 2a, which presents the
TrailM Server architecture). The TrailServices provides basic services to manage the trails and, moreover,
enables applications building their own specialized services according to their needs.
The basic services perform generic tasks which are considered strategic for the management of the server
and the trails. These services are organized into three groups: ManageTrail, ComposeTrail, and QueryTrail.
The ManageTrail group supports the services used for the control and management of the applications. The
ComposeTrail group supports the services used for the trails composition. In this group stand out the service
related to the sending of ptrails in the trailpoint process. The QueryTrail group contains the services that
allow queries related to the trails.
TrailM Server enables applications adding to the model their own specialized services, which can be
composed of basic services. Sect 3.2 describes four specialized services created to support the discovery of
deal opportunities guided by trails. The server has a database to store information of the entities, the trails and
their related content. The entities table stores identification data of the entities related to the trails that are
been managed. The trails table stores the ptrails that compose the entities trails. The trail related content is
stored in the others tables: context, region, event, extension of events and resources (see Figure 3).

Figure 3. TrailM database
Some services can have common functions, for example, to do database query and to manage the trails.
For this reason, we designed the TrailManager component, which performs basic operations that can be used
by the services. The TrailManager aims to meet the basic functionality for trail management as the control of
requests sent to the server, registration and authentication of users, access to the database and, one of its main
tasks, the composition of the record for the storage of a ptrail at the trailpoint solicitation moment. The
TrailRepository provides an interface to access the database repository, thus supporting the TrailManager.
2.2.3 TrailM Client
The TrailM Client resides on the mobile device and supports the applications of TrailM. The client
instantiates and manages the ptrails entities, making the synchronism of applications with TrailM Server. The
client is organized into three layers (Figure 2b).The top layer is the communication (TrailClientComm) client
to the server. The intermediate layer (TrailClientLocation) handles the communication device with the
external provider of geographic location. The management layer (TrailClientCore) manages access to device
IADIS International Conference WWW/Internet 2010
This section presents the strategy used in the deal opportunities discovery guided by trails. Section 3.1
describes the ubiquitous commerce environment, in which TrailM was integrated. Section 3.2 describes how
the trails were used in the discovery process of Deal Opportunities (DOs).
3.1 The Ubiquitous Commerce Environment
The UbiTrade architecture (Franco et al, 2009), presented in Figure 4, is composed of eight components. The
first is the Location System, which allows the identification of the users' current context. The second is the
Profile System that stores the users' data. The third is the Category Tree, which keeps a taxonomy of products
and services. The fourth is the Reference System, which determines the users' reputation, according to the
previous transacted businesses. The fifth is the Service Manager, which operates as a layer of messages
exchange between the interface component and the others. The sixth is the Opportunity Consultant, which is
the main component of the model. It operates as an analysis engine, generating business opportunities from
the profiles and locations of the dealers. Finally, there are two components used as interfaces with users, the
Personal Assistant and the Website.

Figure 4. The UbiTrade Architecture
The deal opportunity identification depends of events occurrence. The opportunities, in UbiTrade, are
classified into three categories. The Buy & Sell category is generated from the desires of a dealer matched
with the offers of another dealer. The Knowledge Exchange category is created based on the desires of a
dealer matched with the desires of another dealer. The Experience Exchange category is produced from the
matching of desires of a dealer with businesses made by another dealer. The identification of the deal
opportunities is made by Opportunity Consultant through the dealers location and profile and the Service
Manager in charge for warning the users about deal opportunities.
3.2 Deal Opportunities Using Trails
3.2.1 Specialized Services in TrailM
The deal opportunities discovery guided by trails was based on specialized services created in TrailM and
accessed by Opportunity Consultant of UbiTrade. Table 1 summarizes the created services. The information
returned by the services is obtained through the analysis of the dealers trails. The preferences are determined
by the use frequency of contexts, regions, and business done in the trail.

ISBN: 978-972-8939-25-0 2010 IADIS
Table 1. Specialized services created in TrailM
Name Description
PreferredContexts Returns a list with the dealers preferred contexts in a trail.
DiscoveryDOs Returns a list containing all deal opportunities of dealers that have a desire and/or offer
in common based on trail.
PreferredDealingContexts Returns a list with the dealers preferred business conclusion contexts in a trail (for
example, jewelry store, restaurant or train).
TypeOfBusinessByRegion Returns a list with the type of business by region based on dealers trails (for example,
3.2.2 The TrailM/UbiTrade Integration
The TrailM/UbiTrade integration was based on the following changes in UbiTrade: (1) we included the
support to the TrailM services in the Personal Assistant; (2) the Opportunity Consultant inferences became to
consider the dealers profiles and the services available by TrailM (described in Table 1).
Figure 5 presents the dynamic of the integration. In UbiTrade, the deal opportunities are warned to the
dealers according to the events occurrence based on Profile System during their trail. Location System
informs the context in which the dealer is to Opportunity Consultant, which identifies the occurrence of one
event that starts the process (step 1). Opportunity Consultant uses this information, combined with dealers
profile and the Reference System, to filter deal opportunities according the reference grade compatible with
the dealer (step 2). For dealers who have the data compatible, Opportunity Consultant matches the profiles,
trying to identify the type of opportunity that they are looking for (step 3). In this stage, UbiTrade uses the
TrailM services to improve its inferences. Identifying an opportunity, Opportunity Consultant sends a
notification to dealers, using the warnings service of the Service Manager component (step 4).

Figure 5. Discovery of DOs guided by trails
This section describes the TrailM and the UbiTrade implementations. Furthermore, it discusses the case study
used to validate the model.
4.1 The TrailM and the UbiTrade Prototypes
The TrailM Server prototype was implemented through the J ava programming language. The Hibernate
framework was used to access the database. The TrailServices layer is based on Web services.
The UbiTrade prototype (Franco et al, 2009) was implemented through the C#programming language.
The Opportunity Consultant and the Reference System components were designed as Windows services.
The Category Tree component and the Profile System were implemented using the Firebird relational
database. The Service Manager was developed using the standard Web service technology. The Personal
Assistant module was developed with .NET Compact Framework and runs on Windows Mobile. Moreover,
IADIS International Conference WWW/Internet 2010
to capture Wi-Fi signals and to store the profiles in the mobile devices, we used the OpenNETCF library and
the .NET System.Xml library, respectively.
The integration between the TrailM and the UbiTrade prototypes was mainly based on changes made in
Opportunity Consultant in order to access the services provided by the Web services of TrailServices. The
TrailM Client prototype was developed through the addition of new functionalities in Personal Assistant.
The UbiTrade location system is based on triangulation of Wi-Fi antennas. Personal Assistant obtains the
antennas powers and forwards to Location System through a Web service. The system uses this information
to infer the mobile device location. The locations must be previously registered on the system database. The
mobile device location is inferred through a comparison between the data sent by Personal Assistant and the
data previously captured through samples.
4.2 Case Study
4.2.1 Obtainment of the Trails
The case study was based on simulated generation of two dealers trails, which were used in the real scenario
described in section 4.2.2.

Figure 6. Relationship of the real and simulation environment maps
Figure 6 presents the relation between the real and simulation environments used in the trails generation,
which is composed of the Houses, the Metro, and the Shopping regions. For example, a movement from
room 215 to room 206 represents the dealers presence on the train, moving from House to Shopping.
We defined these regions because we wanted to monitor the dealers activities while they were moving
from their houses to the shopping. Within the regions, we mapped some location where the users could go,
for instance, the cabstand, the fast food 1, and station 2.
The Metro region represents the area in which the dealers move from their houses to the shopping and
vice versa, therefore, the system can record the events made by the dealer during the movement between the
stations. The third region corresponds to the shopping. Within the shopping region, we defined two
subregions representing the stores and the food court. A region can be composed of one or more subregions,
and each subregion contains at least one location with the values of its geographical position.
Table 2. Example of the records
Dealer Device Event Location Extension Date
Cldio HTC 4351 Entry into location Shopping - Feb. 6 08:03
Cldio HTC 4351 Offer inclusion J ewelry Store Title: Piano
Category: Instrument Sound
Feb. 6 12:01
J oo HTC 4351 Entry into location Shopping - Feb. 6 19:32
Cldio HTC 4351 Exit from location J ewelry Store - Feb. 6 20:03
J oo HTC 4351 Created Opportunity Shopping DO Notification Feb. 6 21:12

In the simulation, the dealers profile information was based in two real people, focusing their preferences
and interests related to desires and offers, and registered in UbiTrade. The trails generation was based on the
ISBN: 978-972-8939-25-0 2010 IADIS
possible events made by the dealers during a period of six months and was simulated using an application
developed specifically for this purpose. Considering possible displacements and deals, the application
generated records related to the events of entry and exit from the locations, offers and desires inclusion,
created opportunities, and performed business. Table 2 presents an example of these records.
4.2.2 Validation through a Scenario
The scientific community has been using scenarios to validate context-aware environments (according to
Deys approach (2001)) and ubiquitous environments (according to Satyanarayanan (2001)). Following this
strategy, we created a scenario for tests and evaluations of the TrailM/UbiTrade integration, involving two
dealers (J oo and Cldio). The scenario was tested in the validation environment described in Sect. 4.2.1. The
dealers, carrying HTC 4351 (see Figure 7(a)), simulated the scenario moving around the environment rooms.
This scenario focuses on the discovery of DOs in different contexts based on the dealers trails, according to
the described below:
Cldio, who lives in Porto Alegre city, wants to sell a piano. During the week he takes the Train to go to
work at the Shopping and have lunch at the food court between 12:15pm and 12:50pm (by the activity
becomes a "preferred context). Joo wants to buy a piano, 3-4 times by month he goes with his family to the
Shopping, sometimes in the afternoon, sometimes at night (even so, the activity characterizes the context as
"preferred"). Therefore, the Opportunity Consultant uses the PreferredContexts service of TrailM to
determine the preferred contexts of each user. Once some preference change is detected, the Opportunity
Consultant uses the DiscoveryDOs service of TrailM to infer possible deal opportunities for dealers. So the
dealer receives a message (see Figure 7(b)) about the deal opportunity and has the possibility to exchange
messages with the counterparty.
Always that the dealers receive a notification of DOs, they can visualize (see Figure 7(c)) or delay the
access. This action indicates that the dealers are interested in the deal, but they do not want to access it at
the moment, i.e., Joo stores the deal opportunity to see in another moment.

(a) HTC 4351 (b) Notification of DO (c) DO details
Figure 7. The scenario snapshots
There are ubiquitous commerce models that use profiles and localization (Shekar et al, 2003; Roussos and
Moussouri, 2004). The iGrocer (Shekar et al, 2003) was designed to be a smart assistant to provide help
while purchasing in supermarkets, its difference is the possibility of tracking and keeping the consumers
nutrition profile, suggesting the purchase of products or even warn the ones that must be avoided.
MyGrocer (Roussos and Moussouri, 2004) is an application projected to assist clients while the purchase
process. However, the main proposal is to control the stock of products. This control is performed in
commercial spots by using sensors on the shelves that will indicate the quantity of such product still left on
determined shelves. In the several works related to ubiquitous commerce (Zhang et al, 2009) none of them
use trails. To our knowledge this is the first work that focuses on the use of trails in ubiquitous commerce. In
addition, the UbiTrade focuses on generating business for the users, whether in the role of customer or
supplier. UbiTrade is also generic, working with the trade of products in different areas, rather than focusing
on a specific operation.
IADIS International Conference WWW/Internet 2010
This article has presented a system to manage trails and its application in the deal opportunities discovery in
ubiquitous commerce environments. TrailM initially has been integrated with UbiTrade, but its proposal is
generic and can be used by other systems. The use of trails allows the systems to be more effective, because
they can use more precise information of the users behavior. The objective of the case study was to
demonstrate that is viable to integrate ubiquitous systems with TrailM as well as to use the trails to improve
automatic decisions, in our case, deal opportunities.
The main conclusions were: (1) the use of trails confirms the mobile computing potential to improve the
deal opportunity discovery; (2) the availability of the history of visited contexts stimulates the use of mobile
devices as instruments to access contextualized contents; (3) TrailM supports content distribution guided by
trails; (4) the prototype and the experiment attest the proposal feasibility.
TrailM as well as its integration with UbiTrade is an initial proposal. The following activities will allow
the continuity of the study: (1) we will run additional tests with the trails obtained from simulation, for
example, the PreferredDealingContexts and TypeOfBusinessByRegion services were not used yet; (2) the
heuristic to determine the preferred contexts (frequency of visit) can be improved through additional
information, for example, time of each use; (3) the services performance involving a significant amount of
records must be evaluated, because the experiment considered a few number.
Baldauf, M., Dustdar, S. , and Rosenberg, F., 2007. A Survey on Context-aware Systems. International Journal of Ad
Hoc and Ubiquitous Computing, Vol. 2, No. 4, pp. 263-277.
Dey, A. et al, 2010. Location-Based Services. IEEE Pervasive Computing, Vol. 9, No. 1, pp. 11-12.
Dey, A., Salber, D. , and Abowd, G., 2001. A conceptual framework and a toolkit for supporting the rapid prototyping of
context-aware application. Human-Computer Interaction, Vol. 16, No. 2, pp. 97-166.
Diaz, A., Merino, P. , and Rivas, F.J ., 2010. Mobile Application Profiling for Connected Mobile Devices. IEEE
Pervasive Computing, Vol. 9, No. 1, pp. 54-61.
Driver, C. and Clarke, S., 2008. An application framework for mobile, context-aware trails. Pervasive and Mobile
Computing, Vol. 4, No. 5, pp. 719-736.
Franco, L. et al, 2009. Exploring Business Opportunities in Ubiquitous Environments. VIII IADIS International
Conference WWW/Internet. Roma, Italy, pp. 469-476.
Hightower, J. and Borriello, G., 2001. Location Systems for Ubiquitous Computing. Computer, Vol. 34, No. 8, pp. 57-66.
Hightower, J ., LaMarca, A. , and Smith, I., 2006. Practical Lessons fromPlace Lab. IEEE Pervasive Computing, Vol. 5,
No. 3, pp. 32-39.
Hoareau, C. and Satoh, I., 2009. Modeling and Processing Information for Context-Aware Computing - A Survey. New
Generation Computing, Vol. 27, No. 3, pp. 177-196.
Levene, M. and Peterson, D., 2002. Trail Record and Ampliative Learning. London: University of London.
Roussos, G. and Moussouri, T., 2004. Consumer perceptions of privacy, security and trust in ubiquitous commerce.
Personal and Ubiquitous Computing, Vol. 8, No. 6, pp. 416-429.
Satyanarayanan, M., 2001. Pervasive computing: vision an challenges. IEEE Personal Communications, Vol. 8, No. 4,
pp. 10-17.
Satyanarayanan, M. et al, 2009. The Case for VM-Based Cloudlets in Mobile Computing. IEEE Pervasive Computing,
Vol. 8, No. 4, pp. 14-23.
Shekar, S., Nair, P. , and Helal, S., 2003. iGrocer: a ubiquitous and pervasive smart grocery. Symposium on Applied
Computing. Melbourne, USA, pp. 645-652.
Smith, A.D. et al, 2006. Towards Truly Ubiquitous Life Annotation. Memories for Life Colloquium. London, pp. 1-2.
Vaughan-Nichols, S.J., 2009. Will Mobile Computing's Future Be Location? Computer, Vol. 42, No. 2, pp. 14-17.
Zhang, L., Liu, Q. , and Li, a.X., 2009. Ubiquitous Commerce: Theories, Technologies, and Applications. Journal of
Networks, Vol. 4, No. 4, pp. 271-278.
ISBN: 978-972-8939-25-0 2010 IADIS
Herli J. de Menezes
, Sean W. M. Siqueira
and Leila Cristina V. de Andrade

Federal University of Rio de Janeiro
Federal University of the State of Rio de Janeiro
Learning objects have been used in e-learning with the objective of allowing reuse content material. However, this
approach also demands getting the right pieces of content and assembling them together in order to build a semantically
richer instructional material. In order to guide this combination of the right content, it is necessary to define sequences of
topics to be covered and then the files to be used as learning objects. In this paper we discuss Learning Object
composition and sequencing from suggestions taken from Knowledge Domain Modeling. We propose the use of a
domain ontology for allowing the suggestion of sequences of topics to be covered, which is also based on some concepts
of Semiotics and Linguistics applied to multimedia narrative structures. Learning environment and the teacher action is
represented as an enactment where educational resources have a role similar to actors in a play. So, the unfolding of a
lesson is conceived as a narrative, like storytelling, so that the elements are assembled in a discourse. A representation of
the Knowledge Domain is coined as ontology represented by means of OWL DL and the relevant concepts are arranged
in a list extracted from the representative graph of the domain. From this list of topics, the user can chose the most
adequate corresponding learning objects to be used in a class or learning scenario. The proposal was validated with
domain specialists and the results are also discussed.
Learning Objects. E-Learning. Learning Objects composition. Ontology. Semiotics.
The web nowadays has much information about almost everything. Finding the right information is not a
hard task anymore. This great amount of available content could be used in educational scenarios. Although
it is somewhat easy to find material about lots of subjects, organizing content to a specific audience is not so
straightforward. It is necessary to make the semantic connections about the concepts to be explored and their
level of explanation. When dealing with learning content there is a common sense towards the reuse
approach, resulting on the so called Learning Objects (LOs). LOs are a kind of resources defined as "any
entity, digital or non-digital, that may be used for learning, education or training" (IEEE, 2002). This wide
definition has the main purpose of encompassing different sort of resources (Wiley, 2002). In this paper, we
are focusing on learning content that is available through the learning management systems or through the
web, so only Digital Learning Objects (DLOs) are addressed. There are several types of DLO such as text
files, diagrams, drawings, mathematical formulas and expressions, exercises (quizzes, problems and
examples), sound and video files and multimedia files.
Rodrigues et al. (2008) stated that creating DLO's is time consuming and expensive, so reusing already
existing DLOs is desirable. In order to make this reuse possible it is necessary to provide mechanisms for
finding and accessing DLOs, and composing them according to the different possibilities of exploring
content or different learning scenarios.
For enabling searching and retrieving DLOs, they are described through metadata elements. There some
metadata standards for describing DLOs, but IEEE LOM (IEEE LTSC, 2002) is the most used. Although
LOM has a wide acceptance, some learning environments use other metadata not completely compatible with
LOM, making finding DLOs a more difficult task as it is necessary to deal with heterogeneity. Therefore,
IADIS International Conference WWW/Internet 2010

composing and sequencing DLO become even more complicated. In addition, DLO composition is
influenced by DLO granularity as Wiley (2002) stated, reusability is in an inverse relationship to granularity.
There are some proposals for composing learning content in the literature. Reusability of DLOs is
discussed in (Barritt, 1999) from a modular perspective where five types are presented: concepts, facts,
procedures, processes and principles. Guidelines are proposed to build Reusable Learning Objects (RLO) by
aggregating (composing) such types. However, there is no guidance on how to sequence the adequate types
in specific learning scenarios and how each type complements the others.
Aroyo (2003) proposed a modularized concept-based framework to build courseware centered on topics
which are conceived as reification of concepts previously classified by an ontology so that its relationships
are used to propose a sequence of topics. However, a contextual approach for topics composition is not
Rodrigues (2008), from a component based approach, suggested a way for LO composition. A flexible
mode of LO composition is proposed using lower granularity components and an algorithm which defines a
semiautomatic composition language based on some learning strategy. Since no specific learning theory is
addressed, this is a general proposal that can be applied in different learning contexts.
In this paper, we consider learning objects composition starting from the conceptual domain modeling
exploring the possibilities of LOs composition built from inferences of the conceptual structure of the
considered domain. We also consider some ideas rooted in semiotic concepts about signs and discourses
generated from the rules for composing narrative structures. As a result we devise some criteria and a
suggestion to compose Learning Objects.
DLO can be considered as a digitally coded physical media supporting some content so it has the main
characteristics of semiotic signs. From this assumption, we can think of a DLO as an entity comprising the
content and the expression aspect. Content is associated to the semantics of the resource and expression to its
physical form.
Some learning-teaching requirements must to be considered when designing a course or a lesson. The
first one is setting the learning objectives which comprises the subject to be taught pertaining to a knowledge
domain (KD). This subject has an internal structure, determined by the main concepts, axioms, experimental
facts and its inter-relationships. From a teaching-learning perspective, this subject must be modeled, so that
the main concepts and their relationships are identified and organized in a coherent way according to the
learning purpose. The second requirement is related to the learning outcomes to be accomplished. These
outcomes are verbal information, intellectual skills, cognitive strategies, attitudes and motor skills (Gagn,
1974). The last requirement is related to the composition itself dealing with the promotion and holding the
learner attention by supplying elements to contextualize its previous knowledge and enhance its curiosity.
These requirements can be associated to the content plane.
The teaching-learning setting is modeled in (IMS, 2003) as a theatrical play, having acts so that each of
them has one or more role-parts. These acts follow a sequence and the role-parts are associated to a role in an
activity. This activity describing the role action in a given environment is, according to the model, equivalent
to an script which defines when the role-parts in an act runs in parallel, when there are many of them, or not.
A lesson could be envisioned as having some features of a narrative structure, as discussed in (Barthes,
1975), (Hagood, 2009), (Chatman, 1975) and (Dijk, 1975). The first is the story that according to (Hagood,
2009), considering the multimedia context, is "collection of experiences represented as resources" and a
discourse representing "what part of the story" is to be told and ''how it is told" (presentation). So this
discourse is just the way the scripts are arranged, or how the story is told.
We claim that for a virtual learning environment (VLE) these features are just like a storytelling system
using LOs acting as resources or characters in a play as defined by (IMS, 2003) so that authoring tasks or
learning design procedures define the way to orchestrate these elements to compose a narrative structure that
holds the attention of its learners and meet the defined learning objectives. We consider some methods for
achieving this from the starting point of a KD modeling in order to define the concept network issued from
this model. KD modeling is done by means of ontology, defined by (Gruber, 1993) as a formalization of a
conceptual model. The obtained concept model can be represented as a graph of nodes corresponding to
concepts and arcs to their relationships. The relevance of a node can be defined by the number of connections
on it (its degree) (Bollobs, 1998). The learning path in this representation corresponds to a path in the graph
connecting two concepts.
The remainder of this paper has the following structure. Section 2 summarizes LOM metadata structures
used for describing LOs. Section 3 presents some features of signs from a structural semiotic perspective.
ISBN: 978-972-8939-25-0 2010 IADIS
Signifier and significant aspects of signs, as well as, Peirces conception of signs are briefly discussed as well
as compositional aspects of signs in a discourse and semiotic features of LOs are stressed. Section 4 discusses
composition of LOs from their metadata structure and KD aspects. Domain ontology is considered a key
point to LO composition since a semantic structure is defined from the concepts of KD and their relationships
to support contextual composition of LO as well as some general rules for LO orchestration. In Section 5 we
present an example of the proposed approach applied to the field of introductory Physics and the opinion of
Physics teachers on composing these concepts. Finally, in Section 6 we draw some conclusions and address
future works.
LOs are described by a data model usually the LOM (Learning Object Metadata). LOM comprises nine
categories such as 1. General; 2. Lifecycle; 3; Meta-metadata; 4. Technical; 5. Educational; 6. Rights; 7.
Relation; 8. Annotation and 9. Classification. These categories are depicted in Figure 1.
In an extensive study on learning objects metadata, Motelet (2007) remarked that the LOM datatypes can
be organized into four categories: (a) Primitive types (size, format, id, URL, time stamp or date) concerning
to attributes like
: General.Identifier, Technical.duration, Technical.location; (b) Langstring type, used in
elements like Educational.description, General.keyword and General.title, so that information can be given in
different languages; (c) Vocabulary instances: Educational.interactivityType (=active, expositive, and mixed)
and (d) vCard: a standard file format for business electronic cards.
The Educational category is very important for educational purposes since it specifies some relevant
parameters to teaching-learning process as the interactivity type useful to match learning resources to the
modality of lesson (active, expositive, mixed). Another example is the Learning Resource Category, whose
value space indicates several kinds of resources to be used, ranging from exercises to lectures, however there
is no indication about its expressing power.
The sequencing and composing task must integrate some features related to the content and form of these
resources. In order to discuss theses aspects, we shift our approach to the semiotic field.

1 We adopt a general notation form LOM attributes in the form Attibute.sub-attribute.

Figure 1. Diagrammatic representation of LOM-IEEE.
IADIS International Conference WWW/Internet 2010

Signs, as discussed in the field of Semiotics, are twofold entities, according to (Saussure, 1945), composed
by a signified and a signifier represented as two layers. As stated in (Barthes, 1975), signified can be thought
as the content layer and the signifier as the expression layer. A further unfolding of these layers was proposed
by (Hjelmslev, 1974) into two sub layers of form and substance. Table 1 schematically shows the structure of
signs, as discussed.
For instance, let us take the sign Mountain: (a) its signifier is the string chain [m,o,u,n,t,a,i,n], (b) its
signified (the mental image of a mountain). The form of expression is given by the phonetic, orthographic
rules to compose the sounds or the characters. The substance of expression is the elements used to compose
(the text font family or the sound waves that are emitted). From the content layer side, the form of content
sublayer is the formal organization of signifiers and the substance of content is the emotional or ideological
aspects, (that mountain I have climbed on the last week-end walk.)
Table 1. Structure of Sign according to Saussure and Hejlmslev
[Adapted from (Chandler, 2001) and (May, 2007)]
Substance Form
Signifier (Expression)
Physical Materials or the medium (words,
sound recorded, words printed)
Language, formal syntactic structure,
technique and style
Signified (Content)
Emotions, Ideological aspects, specific
information content
Semantic structure, Thematic structure,
Conceptual Structures, ontologies

Peirce (1955) represented signs as a triadic relation instead of dyadic representation of Saussure. In his
model, the form which the sign takes is defined as the representamen; the interpretant is the sense made by
the sign and, the object is the substantial aspect of sign. It must be remarked that the representamen is itself a
sign, so, as stressed by (Eco, 1976), a sign is interpreted by means of other signs, in an unending recursive
process, defined as semeiosis (Peirce, 1955). This structure is represented in Figure 2.

Figure 2. Sign components according to Peirce and its interaction (Semeiosis)
A further aspect to be noted is that there are different kinds of signs. Iconic signs stands for an object by
resembling or imitating (Yeh, 2004) while the index refers to the effect produced by the object and the
symbol, which refers to object as the virtue of law, convention or rule. These signs are combined to express
and form a discourse.
Signs composition according to Structural Linguistics occurs in a plane defined by two dimensions. The
syntagmatic dimension is composed by syntagmas, or the string of words that compose a larger unit in the
speech or in a sentence (Saussure, 1945). The syntagma gets its value as opposed to its predecessor or
successor. The second dimension is the paradigmatic one, referred to the associative fields whose
determination can be, in linguistic context, by sound affinity (materialization/conceptualization) or by sense
(education/learning). The speaker chooses from his/her repertory the sign he/she must use to convey the
intended message.
ISBN: 978-972-8939-25-0 2010 IADIS
The outcome of this operation is a discourse defined, in the linguistic field, as a ... formal discussion of a
subject in speech or writing
. In a broader sense, this definition can be extended as a ...system of
representation consisting of a set of representational codes for constructing and maintaining particular
forms of reality within the ontological domain defined as relevant to its concerns (Clandler, 2001).
These semiotic categories can be used to analyze other kinds of 'discourses', such as movies or
hypermedia. In movies domain, we have syntagmatic elements such as long shots, close up, travelings
are matched with sound (special effects, music and dialogues), that are arranged to transmit the film maker or
author's intention according to some paradigmatic composition
. Lets us move to LO field in order to discuss
some of its semiotic aspects.
When a user interacts with a learning environment (LE) he/she establishes a communication link since the
users (receptor/leaner) and a sender (LE) interchange messages. There is, surely, another virtual element,
the author, represented by the LE. Messages are structured as a discourse composed by syntagmas and
structured according to some system, or paradigm. This is not a linear discourse since it comprises different
medias and different and simultaneous codes.
A Knowledge Domain (KD) can be represented by means of ontologies which are the formal
representation of the concepts and relationships on a domain (conceptualization) (Gruber, 1993). Ontologies
can are represented as assertions expressed by means Description Logic axioms. These assertions can be
represented in a machine understanding form by means of OWL DL, a subset of the family of XML-based
languages proposed by W3C consortium (Antoniou, 2008).
The abstract form of ontology is provided by Description Logics, whose main constructs are the set of
axioms, terminology elements and facts. Let us consider, for instance, a set of concepts C and a binary
relationship dependsOn, represented by D. The dependence relationship between the concepts C
and C
conceived as classes of individuals, is expressed in Description Logic as , the set of instances (or
individuals) of C
that depends on the instances of C
. The relationship between concepts can be represented
as a graph, where nodes represent the concepts and edges, the relationship between them. Then, it is possible
to retrieve paths from a concept to another, which is equivalent to exploring the semantic relationships of
learning content to guide the LOs composition.
We can apply these concepts to model a small knowledge domain with only one type of relation. In order
to do this we built an ontology for the Introductory Physics sub domain in order to do a preliminary test for
the proposed method.
From the considered sub domain, we choose a set composed by the concepts: Mass, Force, Work, Velocity,
Derivative, Squared Velocity (SquaredVelocity), Acceleration, Conservative Force (ConsForce), Limit,
Potential Energy, Kinetic Energy and Energy. An Ontology for this set was constructed as OWL DL classes
using Protg 4.1
editor, a free open source editor and knowledge base framework and its graphical
representation using the OntoGraph plug-in is displayed in Figure 3, where the dotted arrows mean the
relation DependsOn and continuous ones stand for the inferred relationship hasSubClass. In order to
accomplish the requirements of the proposed approach we have also defined some relationships between LOs
and concepts using the LOM structures mapping them to the ontology.

2 Random House, Webster's Everyday Dictionary.
3 Travelling refers to the camera movement following the actor walking keeping the same relative position (Source:
4 According to (Metz, 1972) paradigmatic features in a semiology of movies is somewhat feeble, since the inventory of probable units is
open, although finite.
IADIS International Conference WWW/Internet 2010

From the ontology model, a graph structure was modeled (Figure 4) and the set of nodes and edges was
processed in order to determine all the paths in the graph starting from energy node to limit node using a
Breadth First Search algorithm written in Python language. These paths were determined and a concept
dependence list (output list) below was obtained.

Figure 4. Graph obtained from ontology information
>>> find_all_paths(graph,'energy','limit')
['energy', 'force', 'mass', 'acceleration', 'velocity', 'deriv', 'limit'],
['energy', 'force', 'mass', 'velocity', 'deriv', 'limit'],
['energy', 'work', 'force', 'mass', 'acceleration', 'velocity', 'deriv',
['energy', 'work', 'force', 'mass', 'velocity', 'deriv', 'limit']

Figure 3. Proposed OWL DL graph for concepts, learning objects and their relationships
ISBN: 978-972-8939-25-0 2010 IADIS
Notice the results of the output list should be considered in inverted order, i.e., it is necessary to depart
from limit and finish on energy. This output list could suggest a sequence of concepts so that a user could
select matching learning objects from LOM description (keyword) to each concept. From a semiotic
approach, this implies that the composition and sequencing is done, so that the path extracted from ontology
graph is the form of content.
We took the longest path ['energy', 'work', 'force', 'mass', 'acceleration', 'velocity', 'deriv', 'limit'] and asked
them to put the concepts in the order they think is the best way to be presented to the students. The results are
shown in the Table 2.
Table 2. Suggestions for sequencing learning objects as tested by physics teacher
Professor Energy Work Acceleration Velocity Limit Force Derivative
7 6 4 3 1 5 2
7 6 4 3 1 5 2
6 7 4 3 1 5 2
7 6 4 3 1 5 2

It is also important to notice that the longest path ['energy', 'work', 'force', 'mass', 'acceleration', 'velocity',
'deriv', 'limit'] is really similar to the professors answer ['energy', 'work', 'force', 'acceleration', 'velocity',
'deriv', 'limit'], which demonstrates the proposed approach has a good result. A remarkable fact is that all the
professors defined the same sequence and none included the concept mass. However, this fact does not
seems to be of capital importance, since at graduate level, mass is a fundamental concept related to the
inertial properties of the particle. Its deep meaning is worth discussing if gravitational or inertial properties
would be discussed. The other paths were also considered valid by the professors although which is the best
alternative depends on the learning situation. In addition, as the discussion on which were the most adequate
LOs for presenting each concept was divergent, we decided not to address such questions, but to investigate
possible reasons as future works
In this paper we discussed some features of Learning Objects composition and sequencing assuming that
these objects have some properties or characteristics similar to signs as studied in Semiotics. As an
arrangement of signs, this composition shares some features with narrative structures, so that its form of
content is given by the ontology of the addressed KD.
The ontology was modeled from the dependence relationships between concepts. Since ontologies can be
represented as directed graphs, paths can be defined in these graphs so that they can be interpreted as possible
solutions for sequencing learning content to be explored in a class. For each concept in the path, some
suitable set of LOs can be assigned according to a specific learning strategy. This association between LOs
and the concepts can be achieved by means of LOs descriptors that can be used to tentatively map to the most
adequate set of LO to explore some concepts according to the learning situation (learners characteristics,
didactical and pedagogical aspects, technological restrictions etc.). Educational descriptors such as
Interactivity Type, LearningResourceType, InteractivityLevel, and SemanticDensity as well as the physical
aspects (Format, Size, Location, Operating System or Browser requirements, InstallationRemarks and
Duration) are media centric, and generally are closed packed, which demands additional work to decompose
and reuse them.
An example was developed and the evaluation of the approach showed promising results. The ontology
developed for testing the approach aimed to represent the concepts and their relationships for an Elementary
Classical Mechanics sub domain. Four professors were considered to validate the sequence generated by the
However, our proposal has some limitations. The first one comes from the fact that the selection of LOs is
executed manually, however, from the use Semantic Web tools, such as SPARQL query language, selection
could be done by means of the agent technology executing on LOM expressed in OWL DL. The second
limitation is due to the fact that our model is a simplified one, do not encompassing other intrinsic
relationships among the concepts of a complex domain as Physics. The third limitation comes from the
IADIS International Conference WWW/Internet 2010

reduced scope of the test scenario, whose main result was to confirm, in some aspects, the correctness of the
proposed ontology from the teacher practice standpoint. These limitations are going to be addressed in future
It is important to emphasize, however, the quality of the sequencing strategy as it was equivalent to the
professors answer. This is a promising approach to guide students to automatically find the concepts that are
necessary in order to learn a specific subject. Additional work is necessary to automatically make a complete
sequence of learning material according to the learning situation.
Antoniou, G. et al. 2008. A Semantic Web Primer. The MIT Press, New York.
Aroyo, L., et al. Using Topic Maps for eLearning. In Proceedings of The IASTED International Conference on
Computers and Advanced Technology in Education, CATE 2003.
Barritt, C. et al, 1999. Cisco Systems Rusable lnformation Object Strategy. Definition, Creation Overview and
Guidelines. Version 3.0, June 25, 1999. Cisco Systems Inc.
Barthes, R. lements de Smiologie. Paris: ditions du Seuil, 1964. Portuguese translation by Blikstein, r. S Paulo:
Editora Cultrix, 1974.
Barthes, R. et al. An Introduction to the Structural Analysis of Narrative. New Literary History, VoI. 6, No. 2, On
Narrative and Narratives (Winter, 1975).
Bollobs, B. 1998. Modern Graph Theory. Springer Verlag, Berlin
Chandler, D. 2001. Semiotics for Beginners. Available, retrieved in June,
Chatman, S. Toward a Theory of Narrative. New Literary History, VoI. 6, No. 2, 1975.
Dijk, T. A. Action, Action Description, and Narrative. New Literary History, VoI. 6, No. 2, (Winter, 1975).
Eco, U. 1976. A theory of Semiotics. Valentino Bompiani & Co., Milan.
Gagn, R., et al. Principles of Instructional Design. John Wiley & Sons, Inc. New York, 1974.
Gruber, T. Thomas R. Gruber (1993). Toward principles for the design of ontologies used for knowledge sharing. In
International Journal of Human-Computer Studies, Volume 43, Issue 5-6 Nov./Dec. 1995, Pages: 907-928
Hagood, C., et al. Investigating a thematic approach to narrative generation. In: Proceedings of International Workshop
on Dynamic and Adaptive Hypertext: Generic Frameworks, Approaches and Techniques (DAH'09) 2009.
Hjelmslev, L. Prolegmenos a uma teoria da Linguagem. In Ensaios Lingusticos, 91-92. So Paulo: Ed. Perspectiva,
IEEE, 2002. Learning Object Metadata Specification. Available at LOM-Draft.html
IMS. IMS Learning Design, Best Practices and Implementation Guide. IMS Global Learning Consortium, 2003.
Martin, M. Le Langage cinmatographique. Paris: Editions du Cerf, 1990.
May, M. 2007. A Semiotic Framework for the Semantics of Digital Multimedia Learning Objects. In: 14th International
Conference on Image Analysis and Processing Workshops, 2007. ICIAPW 2007, pp. 33-38.
Metz. C. Essays sur la signification au cinma. Paris: Editions Klincksieck, 1968. Portuguese translation A Significao
no Cinema, So Paulo: Ed. Perspectiva, 1972.
Motelet, O. Improving Learning Object Metadata Usage during Lesson Authoring. PhD Thesis. Universidad del Chile,
Peirce, C. S. et al. Phylosophical Wittings of Peirce. Dover. New York, 1955.
Rodrigues, D. S. S. et al., R. N. A Strategy for Achieving Learning Content Repurposing. In: Lecture Notes in Computer
Science. Berlin, Springer- Verlag, 2008.
Saussure, F. Curso de Lingistica Estrutural, p. 91. Editorial Losada. Buenos Aires: 1945.
Wiley, D. et al. The Instructional Use of Learning Objects. Agency for Instructional Technology. Association for
Educational Communications & Technology, 2002.
Yeh et al. 2004. Towards a Semiotic Framework for Using Technology in Mathematics Education: The Case of 3D
Geometry. International Conference on Computers in Education
ISBN: 978-972-8939-25-0 2010 IADIS
Douglas O. de Freitas, Marcos A. R. Silva, Johana M. R. Villena, Bruno A. Sugiyama,
Gilberto Astolfi and Junia C. Anacleto
Federal University of So Carlos
Washington Luis KM 235, So Carlos, SP, Brazil
In this paper, we describe a strategy to improve the Minskys model in order to be used in the search for analogies. We
investigated the Gentners theory to understand how the process of finding analogy is. We realized that the relations used
by Gentner are flexible and dynamic. Thus, we created other kind of relation that we named as meta-relation which is
obtained through Minskys model. We believe that this new kind of relation is more meaningful because it expresses
action and characteristic which, according to Gentner, both attribute more semantic to relation. We made two
experiments to search for analogies in the commonsense semantic network that uses Minskys model: the first one used
just this model while in the second the meta-relations were added. Through these experiments, we observed that the use
of meta-relation increased the quality of the results.
e-Learning, Analogies, Minskys Model, Gentners Theory, Meaningful Learning.
The research and the use of semantics in computer applications is a subject that has been gaining prominence.
Assigning meaning to information or understanding data and knowing how to manipulate them are important
factors for the most effective use of the computer (Liu and Singh, 2004).
One way to enable the effective use of semantics, especially for a better understanding of information by
people, is to understand how people see some information, how they know it, perceive it and believe in it.
Because of that, it is important to find ways to collect, store and process cultural knowledge. An example is
the semantic network from the OMCS-Br project that uses cultural knowledge to develop cultural sensitive
computer applications taking into consideration the users cultural background. Through this semantic
network, it is possible to obtain some resources that can support on creating these contextualized
applications, such as analogy.
A practical use is in educational application where contextualizing it is one of the main pedagogical
issues found in the literature and commonsense knowledge can be used for this purpose (Ausubel, 1968;
Freinet, 1993; Freire, 1996). Ausubels Theory defends a similar approach. According to Ausubel, in order to
promote a meaningful learning, teachers should make the students attach the new piece of knowledge to the
knowledge which is already in their cognitive structure.
Analogy is an example of a way that allows the teacher presenting a new subject according to the
students cultural knowledge. The teacher can explain through analogy new concepts considering other
concepts that the student already know. This facilitates the learning and the memorization. For instance, in
nutrition area, the teacher can explain about a Brazilian fruit called Jabuticaba building an analogy with a
grape: both of them are fruit, small, sweet and round. These characteristics are part of the students common
sense, thus it is important that teachers are aware of this kind of knowledge in order to explain a new one
based on the ones that were already assimilated.
IADIS International Conference WWW/Internet 2010
In this context, the main goal of this work is to improve a commonsense semantic network in order to
perform better searches for analogies and to create meaningful analogies. We intend, through this strategy, to
help teachers promoting meaningful learning.
The paper is structured as follows: the first section shows an overview of the OMCS-Br Project that we
adopt its commonsense semantic network that use Minskys model. The second one describes Gentners
theory about analogy. The following section provides the experiments and our strategies to improve the
search for analogies. The last section shows some conclusions of this work.
1.1 OMCS-Br Project
The Open Mind Common Sense in Brazil (OMCS-Br) project, developed by the Advanced Interaction
Laboratory (LIA) at Federal University of So Carlos (UFSCar) in collaboration with Media Lab from
Massachusetts Institute Technology (MIT), has been collected common sense of general public through a
web site which can be accessed by anyone through The database is
accessible over the internet through an API for computer applications. This common sense database encodes
the relationships between statements generated by people though the common sense database project and
represents cultural knowledge about specific subject. Of particular relevance to our project is that the
database provides demographics that allow us to access the beliefs, attitudes and vocabulary specific to a
particular age group in a particular geographic location.
The website collects data from its visitors. After entering, a person can register and have access to various
activities and themes available for them on this website in order to complete statements about the themes.
Some examples of themes are Personal Preferences, Slang, Children Universe, Sexuality, Health and others.
Statements are completed by filling in a template sentence pattern such as in Figure 1.
Figure 1. Example of two templates where the dynamic part is filled in by the computer, the fixed part is the query
structure and the last part is where the user fills in their thoughts
Words or phrases are typed in by the users in natural language, parsed and stored. The parser derives
tokens from the text and then interprets them using an engine that is based on Marvin Minskys model
(Minsky, 1987) about knowledge representation. This engine generates a semantic network called
ConceptNet (Liu and Singh, 2004), shown in Figure 2.
The ConceptNet organizes the knowledge in a graph where nodes are the concepts and the directional
links represent relations. In this paper, the relations based on Minskys model were defined as Minskys
relation by us. For example, the template banana is used for eat (Figure 1) has a link labeled UsedFor
between banana and eat. This template is stored as a tuple: {UsedFor banana eat}. There are others
relations, such as: IsA, MotivationOf, DefinedAs, CapableOf, etc. as defined by Minsky (Liu and Singh,
2004). His model has shown to be useful for culturally sensitive software development (Singh, 2002).

Figure 2. Example of a semantic network
ISBN: 978-972-8939-25-0 2010 IADIS
This ConceptNet intends to reflect a simplistic, usable structure related to human cognitive structure. It
can support people, through the information encoded, to plan, to execute and to complete an activity. For our
purposes, it is not important whether the facts are true or not that are encoded in the database; in fact, they
will likely be inconsistent as it is the perceptions and beliefs which is what we are after.
Gentner worked with Natural Language Processing (NLP) to identify methods to discover literal similarity,
analogies, metaphors and abstraction (Gentner, 1983). The base that she works is different from OMCS-Br
base which uses Minskys relations to connect and store concepts. The main challenge that we consider is to
understand Gentners theory and map it to be applied in order to expand the semantic network of the OMCS-
Br project to improve analogies search.
Analogy is a comparison between source domain and target domain relations, i.e. edges (relations)
(Gentner 1983). For example, it is shown in Figure 3 one example that illustrates an analogy between Earth
and Electron because there is the same relation (revolve) in both independently of the others concepts. In
other words, Earth and Electron revolve around the sun and the nucleus, respectively.
Figure 3. Two analogy concepts
Minsky and Gentner use different relations to map the knowledge. Gentner uses actions and properties to
compose relations. Therefore, the number of relations are bigger and more specific than Minskys one, which
contain approximately 20 relation (Liu and Singh, 2004), augmenting the concept semantics.
In next section, we present the first experiment using Gentner theory applied in OMCS-Br base.
3.1 First Experiment
We developed a prototype to investigate analogies in the OMCS-Br semantic network. We choose randomly
a concept that we want to look for analogies, in this case the concept banana. Through this concept the
prototype generated a specific semantic network related to the concept banana as shown in Table 1:
Table 1. Specific semantic network of the concept banana
Relation(Concept1, Concept2)
Formal Specification
LocationOf(banana, fruit bowl) R1 (X, Y1)
UsedFor(banana, eat) R2 (X, Y2)
IsA(banana, fruit) R3 (X, Y3)
PropertyOf(banana, yellow) R4 (X, Y4)
MadeOf(ice cream, banana) R5 (Y5, X)
MadeOf(cake, banana) R6 (Y6, X)

Table 1 shows just the first six results as example. With the result of Table 1, the prototype search for
different concepts that match just each relation R in the whole semantic network of the OMCS-Br project.
For each relation, in Table 1, we will show three matches in Table 2:
With these matches we can define a score which represents the times that a concept is linked through the
same relation (R). For example, the score between banana and legume is 1, because they have the
relation LocationOf in common. Table 2 shows the concepts with higher score based in the results of Table 1.
IADIS International Conference WWW/Internet 2010
It is important to observe that the concept that has a higher score is more similar to the concept banana,
i.e. more analogous.
Table 2. Matches for analogy between relations related to banana with other concepts
First Search:
LocationOf(banana, fruit bowl)
Second Search:
UsedFor(banana, eat)
LocationOf(car, gas station)
LocationOf(paper, library)
LocationOf(legume, refrigerator)
UsedFor(doll, play)
UsedFor(computer, work)
UsedFor(pencil, write)
Third Search:
IsA(banana, fruit)
Fourth Search:
PropertyOf(banana, yellow)
IsA(car, vehicle)
IsA(dog, animal)
IsA(woman, human)
PropertyOf(woman, pretty)
PropertyOf(cat, cute)
PropertyOf(fruit, sweet)
Fifth Search:
MadeOf(ice cream, banana)

MadeOf(cheese, milk)
MadeOf(bacon, pig)
MadeOf(table, wood)

Taking into consideration the matches of Table 2, the prototype gives us the analogies concepts based in
Minskys relation. Some examples are illustrated in Table 3.
Table 3. Analogous concepts related to banana order by Score
Concept Score Number of Search (Table 2)
Car 2 first and third
Woman 2 third and fourth
Dog 1 third
Legume 1 First

Observing the results of Table 3, we can see that the concepts are analogous just because they are linked
through the same Minskys relations. For example, in the Table 3, we can observe that the concept car was
considered analogy because it has two same relations (LocationOf and IsA) but these relations are not
specific, i.e., there are a lot of concepts in semantic network with them although we do not identify analogy
among the concepts. It is worth saying that these results showed us that it is not feasible to consider any
Minskys relation to generate analogies. Thus, things like car, dog and woman are considered by this
prototype analogous to banana. On the other hand, we did not consider that a good result. The kind of
relations used by Gentner are more specific and give a stronger semantic. In order to allow that Minskys
relation could have these characteristics, we propose a strategy that combines Gentners theory and Minskys
model to expand its potential. This strategy that we named as meta-relation is described in the next section.
The relations used in Gentners theory are flexible and have a semantic that can be an action or a
characteristic, while Minskys relations are established, approximately 20 (Liu and Singh, 2004). In order to
enhance the potential of Minskys model, we create a strategy to combine his relations to generate more
flexible and specific ones named meta-relations.
The meta-relations have two properties: flexibility and specificity. Flexibility because the meta-relations
are created from the set of relations and concepts in semantic network and when it increases, new meta-
relations can be generated. Specificity because in the meta-relation there are the same characteristics as the
ones Gentner uses, such as action and/or characteristic.
One way to create a meta-relation is combine the relations CapableOf and UsedFor from Minskys
relations, represented as follow:
ISBN: 978-972-8939-25-0 2010 IADIS

The example of meta-relation that we defined is represented by Y, in the formula above. To create it we
combine two Minskys relations (CapableOf and UsedFor) because we perceive that they complement each
other. If X
is capable of Y and if X
is used for Y, then X
and X
can be connected through the action
represented by Y.
For example, to create a meta-relation to the sentence a monkey eats banana, considering the sentences
presented in Figure 1, there is the possibility of creating the meta-relation eat. The mapping to define a
meta-relation is described as follow:
CapableOf(monkey, eat) + UsedFor(banana, eat) = eat(monkey, banana)

The complementation cited above can be observed, in this case, because CapableOf express the ability of
a concept and UsedFor express for what a concept is used for. In other words, if the monkey is capable of eat
and the banana is used for eat, then it is possible to create the relation that express the monkey eats banana,
i.e. eat(monkey, banana).
Thus, we can observe that the meta-relation eat is more specific than the Minskys relations, because it
express an action that, according to Gentners theory, give us more semantic. Different from Minskys
relation, where different concepts can be linked through the same relation, the meta-relation eat involves a
domain of concepts more specific (only the ones involved with this action).
In order to observe the potential of the meta-relation, which apparently has more semantic than Minskys
relation, we decided to make the second experiment in order to investigate if the use of the meta-relation can
improve the analogies results.
4.1 Second Experiment
We choose the same concept used in the first experiment, banana. Through this concept the prototype
generated a specific semantic network related to the concept banana, as shown in Table 4:
Table 4. Specific semantic network with meta-relations of the concept banana
Relation(Concept1, Concept2) Formal Specification
LocationOf(banana, fruit bowl) R1 (X, Y1)
UsedFor(banana, eat) R2 (X, Y2)
IsA(banana, fruit) R3 (X, Y3)
PropertyOf(banana, yellow) R4 (X, Y4)
MadeOf(ice cream, banana) R5 (Y5, X)
MadeOf(cake, banana) R6 (Y6, X)
Cook(mother, banana) MR1(Y7, X)
Cut(man, banana) MR2(Y8, X)
Eat(monkey, banana) MR3(Y9, X)
Buy(supermarket, banana) MR4(Y10, X)
Peel(woman, banana) MR5(Y11, X)

Table 4 shows just the first six results considering Minskys relation and the first fifth results considering
meta-relations. With the result of Table 4, the prototype search for different concepts that match just each
relation R and meta-relation MR in the whole semantic network of the OMCS-Br project. For each meta-
relation, in Table 4, we will show three matches in Table 5. We just added the meta-relation matches because
the Minskys relation matches were presented in Table 2.

IADIS International Conference WWW/Internet 2010
Table 5. Matches for analogy between relations related to banana with other concepts
Sixth Search:
Cook(mother, banana)
Seventh Search:
Cut(man, banana)
Cook(chief, pasta)
Cook(grandmother, potato)
Cook(human, pasta)
Cut(knife, meat)
Cut(scissors, vegetable)
Cut(peeler, potato)
Eighth Search:
Eat(monkey, banana)
Ninth Search:
Buy(supermarket, banana)
Eat(rabbit, carrot)
Eat(child, chips)
Eat(dog, meat)
Buy(butchery, meat)
Buy(restaurant, pasta)
Buy(mother, vegetable)
Tenth Search:
Peel(woman, banana)

Peel(knife, carrot)
Peel(peeler, potato)
Peel(woman, potato)

It is worth pointing out that the first concept in each meta-relation presented in Table 5 represents entities
(Y) that can perform an action (MR) in something (X). For instance, the meta-relation eat(monkey, banana),
formally expressed by MR(Y, X), shows that the entity monkey can perform the action eat in the object
banana. With these matches we can define a score which represents the times that the same concept has the
same relation (R) or meta-relation (MR) as banana. Table 7 shows the concepts with higher score based in
the results of Table 2 and Table 5.
Table 6. Literal similarity concepts to banana order by Score
Concept Score Number of Search (Table 2
and 5)
Potato 4 Sixth, seventh tenth and tenth
Pasta 3 Sixth, sixth and ninth
Meat 3 Seventh, eighth and ninth
Vegetable 2 Seventh and ninth
Carrot 2 Eighth and tenth
Car 2 first and third
Woman 2 third and fourth
Dog 1 third
Legume 1 First

Comparing the results between the two experiments, we noted that the analogous concepts which have the
same meta-relations are more related with banana because they are capable of receiving the same action. For
example, banana and potato are analogous concepts because both can receive the actions of cook, cut and
peel, different from the first experiment, where banana and car are analogous concepts because they have
the LocationOf and IsA relation.
Of course other meta-relation can be considered to refine the results in order to augment the specificity.
The results are just one example to illustrate the potential of the meta-relation, i.e. considering more meta-
relation the results will be stronger.
The early results presented show evidences that the Gentners theory to search for analogies in the semantic
network of the OMCS-Br Project was not suitable. Because, for analogy, it is necessary to use a network with
more semantic, provided by the addition of the meta-relation created by the combination of Minskys
It is important to explain that nowadays there are Minskys relations and meta-relations in the OMCS-Br
semantic network in order to search literal similarity and analogies. On the other hand, in this paper, we just
ISBN: 978-972-8939-25-0 2010 IADIS
presented the use of the meta-relation in analogy search because we consider that using only Minskys
relation does not bring expressive results.
Searching analogies using meta-relation showed good results because when there is the same meta-
relation in two or more concepts, they are more analogous than Minskys relation because meta-relation
express action and characteristic which according Gentner both attribute more semantic to relations.
Nowadays, our prototype does analogy search using Minskys relation and meta-relation. Because of that,
the score is the sum of the both results. On the other hand, in our second experiment there was not this sum,
because there were not common concepts between Minskys relation and meta-relation results as shown in
Table 6. Nevertheless, this sum is possible and it helps to attribute more semantic to results.
A weak point of the meta-relations presented in this work lies in its creation. The meta-relations are
created by matches between Minskys relations, so the meta-relation does not cover all types of concepts,
only those who are able to do some kind of action, expressed by the relation CapableOf. For future work, we
intend to investigate even more Minskys relations to create new matches to generate meta-relations.
Finally, the commonsense semantic network of OMCS-Br combined with the Minskys relation and
meta-relation allow a better search for analogies which any educational computational application can use.
For example, there is a computational application called Cognitor (Buzatto et al, 2009) which any educator
can create on-line learning material taking into consideration students commonsense knowledge. This tool
can provide a list of analogous concepts, considering as parameter concept inputted by the teacher. Through
this material, students can easily understand the subject of the material because it is explained considering the
knowledge that already learned.
It is important to mention that through the API any computer application which has access to the internet
can use analogies to build educational materials or any learning content.
We thank FAPESP, CNPq and CAPES for partial financial support to this research. We also thank all the
collaborators of the Brazilian Open Mind Common Sense Project who have been building the common sense
knowledge base considered in this research.
Anacleto, J. C.; Carvalho, A. F. P. De, 2008. Improving Human-Computer Interaction by Developing Culture-sensitive
Applications based on Common Sense Knowledge. Advances in Human-Computer Interaction. Vienna: I-Tech.
Ausubel, D.P., 1968. Educational Psychology: A Cognitive View. Holt, Rinehart and Winston, New York.
Freinet, C., 1993. Education Through Work: A Model for Child Centered Learning. Edwin Mellen Press, New York.
Buzatto, D.; Anacleto, J. C.; Dias, A. L., 2009: Providing Culturally Contextualized Metadata to Promote Sharing and
Reuse of Learning Objects. In: ACM International Conference on Design of Communication (SIGDOC 2009).
Bloomington, Indiana. p. 163-170.
Freire, P., 1996. Pedagogia da Autonomia: Saberes Necessrios Prtica Educativa, 31 ed. Paz e Terra, So Paulo.
Gentner, D., 1983. Structure-mapping: A theoretical framework for analogy. In: Cognitive Science 7(2), (1983): 155-170.
Godoi, M. S., 2007. Uso de conhecimento de senso comum na educao para a gerao de similaridades e analogias.
Master's degree dissertation, UFSCar.
Silva, M. A. R. ; Anacleto, J. C., 2009. Promoting Collaboration through a Culturally Contextualized Narrative Game. In:
Filipe, J.; Cordeiro, J.. (Org.). Enterprise Information Systems - ICEIS 2009. 1 ed. Berlin: Springer-Verlag, v. 1, p.
Liu, H. and Singh, P. 2004. ConceptNet - A Practical Commonsense Reasoning Tool-Kit. BT Technology Journal 22, 4
(Oct. 2004), 211-226.
Minsky, M., 1987. The society of mind, Simon & Schuster.
Singh, P., 2002. The public acquisition of commonsense knowledge. In: AAAI Spring symposium on acquiring (and
using) linguistic (and world) knowledge for information access. Palo Alto, CA
Singh, P.; Barry, B.; Liu, H., 2004. Teaching machines about everyday life. BT Technology Journal, v. 22, n. 4. pp.
IADIS International Conference WWW/Internet 2010
Hideaki Goto
Cyberscience Center, Tohoku University
Aoba-ku, Sendai-shi, 980-8578 Japan
Collaboration of the network and optical character recognition (OCR) has a great potential to extend the applications of
OCR and to open up new vistas in some future services using OCR. We presented the basic ideas of synergetic OCR
system in 2004, and developed a platformcalled WeOCR to realize the web-based OCR services. The number of
accesses has soared since the experimental servers were put into service in 2005. This paper provides an overview of the
current WeOCR system, shows some updated statistics, and describes the knowledge obtained through a feasibility study
on the web-based OCR systemusing WeOCR. The unique survey conveys some useful hints for future development of
network-oriented OCRs, not limited to WeOCR, and potential OCR applications in the Grid/Cloud Computing era.
Web-based OCR, WeOCR, OCRGrid, character recognition, document analysis, Cloud Computing
Collaboration of the network and optical character recognition (OCR) is expected to be quite useful for
making lightweight applications not only for desktop computers but also for small gadgets such as Smart
Phones and PDAs (Personal Digital Assistants). For example, some mobile phones with character recognition
capabilities became available in the market (Koga 2005). Another mobile phone with a real-time English-
J apanese translator was proposed by a J apanese company. Although it has a simple, built-in OCR, it requires
a large language dictionary provided by an external server on the Internet. Another example is wearable
vision system with character recognition capabilities for visually-impaired people (Goto 2009). Due to the
limitations of the hardware, it is very difficult to put a large character-set data and a dictionary for language
processing into such small gadgets or wearable devices. In addition, we cannot use a sophisticated character
recognition method because the processor power is quite limited. Thus, use of network can be beneficial to
various applications.
The network-based architecture has some advantages from the researchers' and developers' points of view
as well. Recent OCR systems are becoming more and more complicated, and the development requires
expertise in various fields of research. Building and studying a complete system has become very difficult for
a researcher or a small group of people. A possible solution is to share software components as web services.
Since the programs run on the server sides, people can provide computing (pattern recognition) powers to
others without having their source codes or executables open (Lucas 2005, Goto 2006).
We presented a new concept of network-oriented OCR systems called Synergetic OCR (Goto 2004), and
proposed a revised version called OCRGrid in 2006 (Goto 2006). Since year 2004, we have been designing a
web-based OCR system, called WeOCR, as a HTTP (HyperText Transfer Protocol)-based implementation of
OCRGrid. The WeOCR system was put into service in 2005. To our knowledge, there was no commercial
web-based OCR service at that time. In addition, WeOCR is currently the only Grid-oriented, cooperative
OCR platform in the world. One of the biggest concerns with web-based OCR is about the privacy of
documents. A simple question arises Do people really want to use a system like that? To answer this
question, we need to investigate the usage of the system.
ISBN: 978-972-8939-25-0 2010 IADIS
In this paper, we provide an overview of the current WeOCR system, show some updated statistics
following our previous survey (Goto 2007), and describe the knowledge obtained through a feasibility study
on the web-based OCR system using WeOCR.
2.1 Basic Concept
We designed OCRGrid to make a lot of OCR engines work cooperatively over the network worldwide to
gain some synergetic effects such as performance improvement of OCR, to realize multilingual, high-
function, sophisticated OCR systems, and to provide ubiquitous OCR services. The basic concept of
OCRGrid is quite simple. Many OCR servers are deployed on a network as shown in Figure 1. The OCR
servers work either independently or cooperatively communicating with each other. A client connects to one
of the OCR servers and sends text images to it. The servers recognize the images and send the recognition
results (i.e. character codes, etc.) back to the client.
Although OCRGrid is expected to be used world-wide over the Internet, we may use any networks such
as wireless networks, corporate local area networks (LANs), virtual private networks (VPNs),
interconnection networks in parallel servers, and even inter-process communication channels.
Some potential applications of the platform have been discussed in (Goto 2004, Goto 2006).
Unlike some commercial OCR systems equipped with web user interfaces, we are expecting WeOCR
servers will be supported by the communities of researchers, developers, and individuals as well as
application service providers.
recognition results
send images
cooperative work

Figure 1. OCRGrid platform
2.2 WeOCR Toolkit and Server Search System
Making a web-based OCR from scratch as an application server requires a lot of expertise about network
programming and network security. Since many OCR developers and researchers are not so familiar with
network programming, we developed a toolkit to help those people build secure web-based OCRs easily.
The document image is sent from the client computer to the WeOCR toolkit via the Apache web server
program. The programs in the toolkit check the image size, examine the file integrity, uncompress the file if
necessary, convert the image file into a common file format, and invoke the OCR software. The toolkit
includes some programs for protecting the web server from malicious attacks or from the defects of the OCR
programs. The recognition results are converted into HTML data and sent back to the client.
To enable the end users to search for appropriate OCR engines easily, we developed a directory-based
server search system. Figure 2 depicts the overview.
IADIS International Conference WWW/Internet 2010

Figure 2. Directory-based server search system
A specification file written by the server administrator is attached to each OCR engine. The specification
file is written in XML (eXtensible Markup Language) so that automated data handling becomes easy. The
file describes the specifications of the server, including the location (URL) of the server, the location of the
application interface program (CGI), the name of the OCR engine used, supported languages, supported
document types, etc. The portal server has a robot program for collecting the specification data automatically
and periodically from the OCR servers. The robot analyzes each specification file in XML and updates the
database entries. A simple search program picks up the OCR servers that match the client's needs from the
database and shows the search results.
2.3 Available Servers
Table 1 shows the WeOCR servers known to the author as of Sept. 2010. All the servers are constructed
based on Open Source OCR software. Although these OCR engines work very well with some limited kinds
of documents, the average performance is basically inferior to that of commercial products. Thanks to
Tesseract OCR (Smith 2007), the situation is improving.
Table 1. WeOCR servers (as of Sept. 2010)

Since there was no Open Source OCR for J apanese, we also developed a J apanese OCR engine called
NHocr from scratch to contribute to the developers and end-users who eager to have one. (See
Unfortunately, there is no server outside our laboratory today. Actually, there was a Turkish WeOCR
server with a private (closed) post-processing in Turkey. It was shut down sometime a couple of years ago
for unknown reasons. We have heard from some developers that they are interested in deploying OCR
servers. However, none has been completed or opened to the public as far as we know. Although some
Server name
/ location Sub-ID Engine Deployed Description
ocrad GNU Ocrad Feb 2005 Western European languages Michelle
/ J apan
(formerly known as
gocr GOCR Nov 2005 Western European languages
ocrad GNU Ocrad J un 2006 Western European languages
hocr HOCR J un 2006 Hebrew OCR
tesseract Tesseract OCR Aug 2006 English OCR
scene Tesseract OCR
+private preprocessor
Apr 2007 Scene text recognition for
English (experimental)
/ J apan
(formerly known
as appsv.ocrgrid)
nhocr NHocr Sep 2008 J apanese OCR
J imbocho
/ J apan
Feb 2010 (mirror of ocrad, tesseract and
nhocr on Maggie)
ISBN: 978-972-8939-25-0 2010 IADIS
commercial web-based OCR services have been launched so far, none of them has WeOCR-compatible Web
API (Application Programming Interface).
The following list shows the web-based OCR services and demos spotted by the author.
Web OCR service by KYODONETWORKS (for J apanese text only)
OCR Demo by Startek Technology (for printed numerals only)
ocrNow! by Wordcraft International Limited (limited trial)
Google Docs by Google (built-in OCR engine can handle some European languages only)
None of them takes into account Grid-based, collaborative use of OCR engines as far as we know.
Our idea of Grid-style open OCR servers has not been successful yet. For commercial providers, one of
the reasons is probably the lack of plausible business model. From the researchers and developers points of
view, the incentives may not be so clear because the return gains are often indirect. Without a lot of
contributions, we would not be able to achieve collaborative environments or synergetic effects such as
multilingual processing. We need to investigate the feasibility of OCRGrid platform further.
3.1 Access Statistics
We are interested in how frequently the WeOCR servers are used. A survey was conducted from Nov. 2005
to J uly 2010. Note that a warning message Do not send any confidential documents is shown on every
image submission page (portal site). Some detailed information about the privacy is also provided.
Figure 3 shows the monthly access counts on the two WeOCR servers, Michelle and Maggie. The
numbers of requests to six engines (except NHocr) are added together. The line access represents the
numbers of requests, while another one image_ok shows the numbers of successful image uploading. The
failures were mainly due to file format mismatch and interrupts of data transmissions.
The number of accesses was increasing gradually in the first year, despite the privacy concerns, until the
first automated client was spotted. Then, the server load began soaring and reached to the current average
around 400,000 counts/month (J uly 2010). One of the heavy users seems to be running a program for
monitoring numerals shown on displays of some kinds of apparatuses. Figure 4 shows the access counts on
the NHocr server only, which is almost free from an automated client. The server load has reached to the
current average 25,000 counts/month.

Figure 3. Monthly process counts. (Michelle +Maggie)
IADIS International Conference WWW/Internet 2010

Figure 4. Monthly process counts. (NHocr on Maggie)
These results show that there are a lot of applications of OCR in which privacy does not matter. Statistics
of processed image types can be found in our previous report (Goto 2007).
3.2 Applications of WeOCR
The WeOCR platform has been used in various ways.
For example, majority logic is known to be useful for combining multiple classifiers and improving the
accuracy of character recognition (Miyao 2004, Tabaru 1998). We also investigated the accuracy
improvement based on the majority logic using WeOCR (OCRGrid) platform (Goto 2006). Some private
OCR engines were deployed on a local WeOCR platform in our laboratory in the experiments. A client
program sends each character image to multiple WeOCR servers, collects the recognition results from the
servers, and chooses the most popular character as the top character candidate. A WeOCR server itself can
also become a client. Since the methods based on majority logic require a lot of OCR engines with different
characteristics, the Grid-based approach should be quite useful.
People in Seeing with Sound The vOICe project (Meijer 1996) have made an e-mail interface for
mobile camera phones. The server receives an image from the mobile phone and sends the recognized text
data back to the user. An OCR engine on the WeOCR platform is used as a back-end server. The system was
originally developed in order to help visually-disabled people read text on signboards, etc. Any OCR
developers can help those people through the WeOCR platform.
Table 2 shows the WeOCR application software/server known to the author so far. These applications use
WeOCR platform instead of a built-in OCR library. All the applications except the last one were made
independently by developers having no special relationship to our laboratory. In addition, some other
experimental WeOCR client programs and web services have been spotted on the Internet. Thus, our
WeOCR platform has been accepted as a useful building block by many developers, and created new OCR
application design and usage styles suitable in the Grid/Cloud Computing era. Actually, any developers can
add OCR functions to their programs quite easily using WeOCR platform.
A potential future application of WeOCR (OCRGrid) platform is multilingual processing. Thousands of
languages exist all over the world. ABBYY FineReader, one of the worlds leading OCR packages, can
handle around 180 languages so far. However, the spell checking is supported for about 40 languages only. It
seems impossible to have a large number of languages and dialects supported by OCR engines from only a
couple of companies. WeOCR platform can provide a lot of community-supported OCR servers with
localized dictionaries. We will be able to obtain better recognition results, since we can expect that the
servers for a language are better maintained in the countries where the language is used. In addition, having a
lot of OCR servers for various languages are very useful for research on multilingual OCR systems.

ISBN: 978-972-8939-25-0 2010 IADIS
Table 2. WeOCR client software. (as of Sept. 2010)
Name Platform Description
The vOICe server (server) e-mail interface for mobile camera phones
WordSnapOCR Android Proof-of-concept application for word input using camera on
Kanji Yomi
Eigo Yomi
Kanji Yomi Lite
Android Kanji/English reader with online Kanji-English dictionary look-up
Cest What? Android, iPhone A mobile OCR & translation application for iPhone and Android
Android Android frontend for WWWJDIC
Reading Assistant
online edition
Windows A proof-of-concept toy Reading Assistant for helping the visually-
impaired (developed by the author)
3.3 Extensions of WeOCR and OCR Engine
During the WeOCR usage observation, we have recognized a need for single-character recognition in
J apanese and Chinese. Looking-up individual Kanji (or Han character) is a popular way of finding the
meaning, stroke order, similar characters, etc. Figure 5 shows an example of single Kanji look-up.
Although single-character recognition mode was taken into account in the early design of WeOCR
toolkit, we have revised the API and defined a new data structure so that client programs can receive some
top candidates of characters. We have also added single-character mode to the OCR engine NHocr to fulfill
the demands in some applications. NHocr produces a few top candidates of characters, and this function
allows the end-user to choose a correctly recognized character.

Figure 5. Character recognition and Kanji look-up by Kanji Yomi
We have introduced WeOCR system, which was designed to realize a distributed, cooperative, synergetic
OCR environment, and described the results of a feasibility study on the web-based OCR system. We have
been operating the worlds first and only Grid-style web-based OCR servers for almost five years. According
to the usage statistics, the number of accesses to the servers has soared significantly. The following
conclusions have been drawn.
There are a lot of applications of OCR in which privacy does not matter. Many people are willing to
use online OCR despite the privacy concerns. Actually, some commercial web-based OCR services have
been launched these couple of years.

IADIS International Conference WWW/Internet 2010
WeOCR system has become a nice building block for various OCR application software / servers in
the Grid/Cloud Computing era. Some interesting applications have been developed and become popular,
especially on Smart Phones. In other words, WeOCR has been successful in expanding OCR applications and
in providing developers with ready-to-use OCR engines.
Full-featured OCR with strong document layout analysis is obviously desired. On the other hand,
even a single-character recognition feature has been proved to be quite useful in various applications.
Despite the open, Grid-oriented design of WeOCR, our laboratory is currently the only service
provider. None of the commercial servers is known to have WeOCR-compatible Web API. Our idea of
Grid-style open OCR servers has not been successful yet.
A popular complaint from the users is about the precision of the OCR engines. As long as we use
Open Source OCR engines, the precision is limited in general. There is an urgent need to provide the users
with better OCR engines online.
If some state-of-the-art OCR engines are registered, WeOCR (OCRGrid) system will be very useful not
only for end-users worldwide but also for researchers and developers working on OCR systems. We would
like to encourage many developers and researchers to provide WeOCR services of their own. Our future
work includes investigating the feasibility of OCRGrid platform further, developing some applications and
architectures to draw synergetic effects, and refinement of WeOCR API and toolkit.
This work was supported by Grants-in-Aid for Scientific Research No.18650036 mainly, and No.22300194
partially, from J SPS, J apan. The author would like to thank all the developers of WeOCR client applications
and services.
Goto, H. and Kaneko, S., 2004. Synergetic OCR : A Framework for Network-oriented Cooperative OCR Systems.
Proceeding Image and Vision Computing New Zealand 2004 (IVCNZ2004), pp.35-40.
Goto, H., 2006. OCRGrid : A Platform for Distributed and Cooperative OCR Systems. Proceedings of 18th International
Conference on Pattern Recognition (ICPR2006), pp.982-985.
Goto, H., 2007. An Overview of the WeOCR Systemand a Survey of its Use. Proceeding Image and Vision Computing
New Zealand 2007 (IVCNZ2007), pp.121-125.
Goto, H. and Tanaka, M., 2009. Text-Tracking Wearable Camera System for the Blind, Proceedings of 10th
International Conference on Document Analysis and Recognition (ICDAR 2009), Volume 1, pp.141-145.
Koga, M. et al, 2005. Camera-based Kanji OCR for Mobile-phones: Practical Issues. Proceedings of 8th International
Conference on Document Analysis and Recognition (ICDAR 2005), pp.635-639.
Meijer, P., 1996. Seeing with Sound The vOICe (website),
Miyao, H. et al, 2004. Printed J apanese Character Recognition Using Multiple Commercial OCRs. Journal of Advanced
Computational Intelligence, Vol.8, pp.200-207.
Smith, R., 2007. An overview of the Tesseract OCR Engine. Proceedings of 9th International Conference on Document
Analysis and Recognition (ICDAR 2007), Volume II, pp.629-633.
Tabaru, H. and Nakano, Y., 1998. A Printed J apanese Recognition SystemUsing Majority Logic. Proceedings of Third
IAPR Workshop on Document Analysis Systems (DAS98), pp.185-188.
ISBN: 978-972-8939-25-0 2010 IADIS
Roland Grlitz, Benjamin Seip, Asarnusch Rashid and Valentin Zacharias
FZI Forschungszentrum Informatik,
Karlsruhe, Germany
Contemporary studies show that the usage of health-related web portals increases continuously. Additionally, the
internets usage is in a transition towards more user participation by allowing more and more user generated content. In
the area of health care this development is commonly called health 2.0; a development that leads more patients and
doctors to actively participate and contribute on health portals their virtual communities. This results in numerous
opportunities for health care providers to capitalize on online health communities.
In order to aid the understanding of this emerging trend, this paper presents an extensive analysis of up-to-date German
web portals that provide health-related content from a user perspective. They are classified as well as clustered according
to their characteristics regarding usage of web 2.0 technologies and health care specialization. With this clustered
overview of the current health care web portals this paper tries to clarify the term health 2.0. Additionally, general design
options for health care portals are drawn from the analyzed web portals and state-of-the-art literature to aid in the
development of new health care portals.
Health 2.0, Medicine 2.0, e-Health, health care, virtual community, review
With emerging new technologies and applications the use of the internet changed drastically in the course of
the past decade giving rise to the term Web 2.0 (O'Reilly, 2007). In the frame of web 2.0, many derivates
have appeared, for example music 2.0, enterprise 2.0, science 2.0, health 2.0 and medicine 2.0. The latter two
are found to be denoted very similar and both terms are generally attributed to eHealth, which already popped
up in the context of telemedicine in the year 2000 (Hughes et al., 2008, Mitchell, 2000). It is used
inconsistently and unfortunately there is no clear definition, but due to the strong influence of the internet, its
conception is gradually shifting to health care practices based upon usage of telecommunication and internet.
Nevertheless, eHealth, in contrast to medicine 2.0 and health 2.0, does not imply collaboratively generating
and utilizing content. In particular, health 2.0 is characterized by an interactive information sharing and user-
centered design. Nevertheless, it is not defined what extent of user-centered design health 2.0 generally
According to contemporary studies, the use of health-related web portals increases continuously. Due to
advancing health 2.0, the user generated content (UCG) is becoming more and more important (Kummervold
et al., 2008). This offers numerous chances, because the doctor is an expert in identifying a disease, but the
patient is an expert in experiencing it. In this context there are many publications that discuss the influence of
the internet on medical practice in general and on individual consumers focusing on information quality, trust
and acceptance. Furthermore, there are studies that focus on social interaction and benefits of crowd sourcing
in health care. However, there are no publications that review and analyze health-related web portals
utilization of web 2.0 technologies and virtual communities, thus the importance of user generated content in
their business model.
This paper presents a broad review of German health-related web portals. The range of incorporated
social media is analyzed from a user perspective and they are clustered according to their degree of
specialization in health care and potential interaction between users. Due to this two-dimensional
IADIS International Conference WWW/Internet 2010
characterization, the websites are clustered according to their purpose, offered services and means of
participation, which facilitates the comparability among the portals. Individual portals are then analyzed with
a focus on the utilization of web 2.0 components and virtual communities. Then a set of design options that
allows describing the web portals business model according to a framework for virtual communities
described by Leimeister et al. (Leimeister and Krcmar, 2004). Thus, by reviewing contemporary health-
related websites, design options for virtual community business models in health care are identified.
In the last decade the internet started to be perceived as an important source of information and naturally it
also became a source for patients seeking health information (Wilkins, 1999, Baker et al., 2003, Diaz et al.,
2002). Since health information are vital, many publications assess the quality of health information on the
internet (Impicciatore et al., 1997, Eysenbach, 2002, Baker et al., 2003, Wilkins, 1999), assess the
trustworthiness of experts performing these quality assessments (Craigie et al., 2002) and evaluate its effect
on transforming the traditional system of medical primary care (Murray et al., 2003, Cross, 2008).
However, with the upcoming of web 2.0 technologies, the research focus has switched to the effects of
the collaborative components of health care websites, which utilize the collective wisdom of the crowd. For
example, Eysenberg discusses the influence of social networks as well as the willingness for patients to
collaborate regarding health related subjects online in his article about medicine 2.0 (Eysenbach, 2008).
Because of an increasing number of people that use social networks regularly, an increasing number of
people use social networks to find answers to health-related problems. Nevertheless, there was no evidence
found that the influence of virtual communities and online peer to peer interaction did have a significant
impact on the patients convalesce (Eysenbach et al., 2004, Griffiths et al., 2009). Furthermore, in
interrelation to the common lack of willingness to pay for content that is available online, people are not
willing to pay for health related content either (Adler, 2006).
The methodology of health portal analysis is two-fold and has a qualitative as well as a quantitative approach.
The first part consists of a quantitative search for German health-care-related portals, their classification and
subsequent clustering. In the second part, a set of business-oriented design options for community-centered
health care portals is identified through qualitatively analyzing particular portals more closely and state-of-
the-art literature.
3.1 Searching, Classifying and Clustering of Health Portals
The search process for the German health-related web portals incorporates three procedures: a web search
using Google, Bing and Yahoo with the German key words depicted in Table 1 in which the first 20 search
results are considered excluding pages that do not target an audience seeking health-related information,
products and services; interviews with healthcare researchers as well as domain experts; and examining
relevant links on the retrieved websites that are identified with the first two procedures.
Table 1. Translated German key terms used for the web search
Care relatives
Care support
Health Care
Health insurance
Helping the elderly
Living in old age
Nursing home
Nursing service
Retirement home
Self-help group

After the search process, the identified health-related web portals have to be classified in order to
compare their characteristics as well as cluster similar portals together. Consequently, the portals are
classified along two dimensions, which assess their accentuation of health care and user participation.
ISBN: 978-972-8939-25-0 2010 IADIS
Precisely these two dimensions were chosen, because they grasp the core of health 2.0: health care and user
The results of the classification are plotted in a chart in which the x-axis denotes the degree of web 2.0,
which represents the user involvement in generating the offered content and the y-axis denotes the websites
degree of specialization in health care. The degree of user involvement ranges from 1, indicating no user
involvement at all, to 5, in which the user provides all the content. The second dimension, the degree of
health care specialization also ranges from 1 to 5, whereas 1 is a rather universal portal that among others
also addresses health care issues (e.g.Wikipedia) and 5 implies a very specialized website focusing on one
aspect of a disease. All the found web portals are classified according to the two introduced dimensions by
two persons independently. If the two analysts classify one particular website differently, it is assessed again
and its characteristics are determined together.
After searching and classifying the German web portals with health-related content, they are exploratively
clustered subject to their position in the chart that indicates the individually classified specialization in health
care and degree of user involvement.
3.2 Identifying Design Options for Health Portals
In the second part of the analysis design options for virtual community business models are determined
exploratively. They are grouped according to a framework describing virtual community that was introduced
by Leimeister et al. (Leimeister and Krcmar, 2004). Leimeister defines four separate models that need to be
considered when designing a virtual community, thus the design questions are grouped according to the
partial models for strategy, products and services, actors and revenue. An initial set of design questions is
specified iteratively through analyzing representative web portals that have been identified in the first part.
4.1 Results of the Health Care Web Portal Classification
The search yielded 246 German web portals with health care related content. The specific URL and the
corresponding degrees of specialization in health care and web 2.0 of each web portal are depicted in Table 2.
Table 2. The 246 web portals and the classified values for their degree of web 2.0 and degree of specialization in health
care denominated X and Y, respectively
Web Portal X Y Web Portal X Y Web Portal X Y 1 4,5 1 3 1 4 3 2 1,5 3 1 4 1 2 1 3 1 4 2,5 5 2,5 2,5 1 4 1 3 1,5 2 4,5 4 5 4,5 1 3 4 3 2 4,5 2,5 2 1 4 2 4,5 5 1 4 4 2 1 1 4 1 4 1 4 5 1 1,5 4 1 1 1 4 1 3 2,5 2,5 2,5 1,5 1,5 4 3 3 2,5 1 1 4,5 1 2,5 1 1,5 2,5 4
1 2,5 1 2 5 4 2,5 2,5 4 3 2,5 4 1 4,5 1 3 3,5 2 1,5 3 1 3 4 2 1,5 3 4 2 1 2 1 3 2 3 2,5 3 1 4 1 1 2 1 1 2,5 1 2,5 2,5 2,5 1,5 2 1 3 1 1
IADIS International Conference WWW/Internet 2010
51 1 1 1 3 1 2 2,5 3 1 3 1 2,5 1 3,5
4,5 4 1 2 4,5 1 1 4 2 5 1,5 5 1 4 1 4 1 4 1 1 2,5 5 1 4 1 2 1 4 1 3 2,5 5 1 2 1 3 1 3,5 2,5 3,5 2 4 2,5 3 1 2 1 1,5 2 4 4 2 2,5 4 2,5 2 2,5 2 1 5 2 2 4 2 1 2,5 2,5 3 1 2 1 2 1 3 1 4 4,5 4 1,5 2 1 4 1 4 2 3 1,5 2 1 2 1 3 1,5 2 2,5 4 1 3 1 2 1 4,5 1,5 3 4 2 2,5 1 3,5 3 3 2 2,5 4,5 1 3 1 1,5 1 4,5 2 4 2,5 1 1 4 1 2 1 3 1 4 1 4 1 3 2,5 5 2,5 2 1 4 1 5 1 2,5 1,5 4
1 3 1 4 1 2,5 1 4 2 4 1 4 4,5 3 2,5 4,5 1 3 1 1,5 2,5 5 1 2,5 2,5 2 1 5 1 1 2,5 5 1 5 4 2 2,5 3 2,5 5 1 3 4 3 4 5 1 3 1 3 4,5 4,5 1 3
2 3 5 1 1 3 2 3 1 2 1 2 1 2 1,5 3 1 2 2,5 3 2,5 4 1 2 1 1,5 1 2,5 1 2 5 1 1 5 1,5 2
Web Portal X Y Web Portal X Y Web Portal X Y 1 1 1 4 1 2 2,5 1 2,5 3 1 2 4,5 4 2,5 2 1 2 1,5 2 1,5 1 1 2 5 1 2,5 2 1 2 2,5 2 4,5 3 2 2 1,5 2 1 3 5 1 1 4,5 1 3 2,5 4 4 2 4,5 3 1 1
1 3 1 3 1 4 1 4 1 3 1,5 3,5 2,5 2 1 3 1 3,5 1 3 1 3,5 1 4 1 2 1 3 4,5 1 1 2 2 4 2,5 3 2,5 2 1,5 4 1 4,5 1 2 1,5 4

In Figure 1 the distribution of the 246 individually assessed web portals is visualized. While determining
each portals characteristic degree of health care specialization as well as the utilization of web 2.0
technologies, it became evident that from a user perspective the offered services and social network
components are difficult to compare. Therefore, the identified web portals are clustered according to their
individual characteristics into 23 exploratively identified clusters, which is illustrated in Figure 2. Since every
web portal can be appointed to a definite cluster, a complete analysis of each individual web portal can be
ISBN: 978-972-8939-25-0 2010 IADIS
avoided by thoroughly analyzing representatives of the clusters. Furthermore, the cumbersome individual
differentiation can be replaced by a sophisticated delineation of the identified clusters characterization.

Figure 1. The distribution of the 246 classified German health care web portals
The 23 identified clusters can be separated into two major groups, whose division is illustrated by the
horizontal line in Figure 2. The 7 clusters below this line contain e-shops, information portals or wikis that
only partly deal with health care, but are rather general. The clusters above the line have at least the key issue
health care. These higher 16 Clusters are roughly divided into 4 categories on each of the two dimensions.
There are pure information websites or e-shops that do not grant users permission to add content at all.
Another group consists of web portals that offer content, which is generated by the operator, but can be
commented by users directly or through an affiliated forum, in which users can freely post content. If the
web portal is completely user-centered and its main focus is the user generated input in a forum, it is in a
community cluster, and if the portal only consists of freely generated user content, it is in a wiki cluster. The
rough division in the health care specialization dimension yields the following four groups: focus on wellness
or senior citizens, which have a high proportion of health care subjects; focus is generally health care,
medicine or physician; focus on one health care aspect or disease groups, e.g. care giving or cancer; focus on
certain health care issues or particular disease, e.g. dementia or multiple sclerosis.
IADIS International Conference WWW/Internet 2010

Figure 2. The identified 23 clusters of the German health care portals
There are numerous examples for information portals that do not allow any user content: for example or in the wellness and senior cluster; for a
physician search engine; for health information; is an
health-related e-shop; for care information; and for information
on the particular disease COPD. Except for e-shops the business models of most of these information portals
are based on providing health-care-related information while being cross-financed or financed through
banner ads.
The afore mentioned clusters also exist with some user involvement in the form of possible comments or
associated forums. Because of stronger user integration, most of the portals in this group of clusters
incorporate different ways of generating revenue in their business model. Apart from banner ads they also
have more tailored sources of revenue. For example is a portal for physicians and medical
employees offering an e-shop, a forum, videos and medical information. There are also additional services,
which are subject to fees like access to medical literature, further education, software and videos.
Furthermore, third parties offer supplementary services like CRM-systems. Thus, the portal has a wide array
of revenue sources, ranging from simple sale and premium accounts to individual services.
Even though in the United States there are several business models that are based on virtual communities
and their user-generated content, for example,, and, German community-centered health care portals are rather
scarce. Their number even decreases with increasing specialization in health care. There are only two portals
that have an innovative business model: and They are in the
clusters care community and patient community, respectively. The care community portal Betreut is an
intermediary for family services. People can offer their service, e.g. taking care of the children, on the portal
and others can search for it. The service seeking person has to pay a fee to establish the contact and can then
rate the received service. Additionally, the portal operator sells licenses to companies for using the platform
commercially. This care community portal combines online and offline contacts while assessing the quality
of the services through user feedback. The patient community MS-Gateway is a portal sponsored by
ISBN: 978-972-8939-25-0 2010 IADIS
BayerHealthCare and only deals with the disease multiple sclerosis. It offers various information, studies and
literature on multiple sclerosis and allows the users to exchange details of their personal health record and to
enter a personal health therapy sponsored by Bayer. The portal is supposedly financially supported by Bayer,
but its business model might be self-sustaining due to the offered personal health therapy.
4.2 Determining Design Option
In addition to the health portal review, a catalogue of 70 design options for virtual community portals in
health care was created. The initial options were iteratively refined by determining their value for exemplary
websites out of the 16 more closely considered clusters. The final set of design options that have an
enormous influence on the business model for health care virtual communities is divided into the four
categories strategy, products and services, actors and revenue. The main aspects covered by the design
options are
- strategic questions, such as the legal form of the portal, its language and regional or national
orientation, the used technology and the decision between commercial or non commercial;
- questions concerning the products and services, i.e. what is offered, how much user generated
content is allowed, is the portal specialized in a certain healthcare domain, how to ensure credibility and what
is the actual added value for the future customers;
- involved participants and their communication, i.e. is communication between users allowed, are
offline contacts between users desirable, how much are users involved in the community as well as the portal
und do I want to incorporate third party content;
- possible revenue, such as connections between the offered services and products, transaction
dependent as well as transaction independent sources of revenue and the question of whether the user or third
party members have to pay.
One striking aspect distilled from the conducted review of German health care web portals is that most of the
found portals are predominantly web 1.0, for which the operator provides and controls all the information
that is published. In contrast to the possibilities user generated content offers, only a fraction of the
investigated portals tries to live on these potentials by applying health 2.0. Furthermore, there are many
different concepts of user generated content, ranging from simply rating the professionally created content up
to creating all the content available on a website. Particularly, portals that deal with specialized health care or
diseases, for example multiple sclerosis or regional care giving support groups, try to employ the most user-
centered designs and interactive information sharing. Thus, a common definition of health 2.0 cannot be
This reviews design options for developing health care community portals offer support for the creation
of a new health care web portal. However, the portals have only been analyzed from the users point of view.
Certain options cannot be defined exactly, because data either from the operator or from other reliable
sources is necessary. Therefore, an interesting future approach would be to interview the stakeholders of
particular health care web portals.
Despite its drawbacks the accumulated and reviewed health care portals as well as the created design
options are a unique snapshot of contemporary German health care portals and a valuable instrument,
respectively. In a further step either specific health care portals that incorporate user generated content need
to be investigated further by obtaining data from the operators or a newly designed web portal will be
prototyped and evaluated.

IADIS International Conference WWW/Internet 2010
Adler, K. G. 2006. Web portals in primary care: an evaluation of patient readiness and willingness to pay for online
services. Journal of medical Internet research, 8, e26.
Baker, L., Wagner, T. H., Singer, S. & Bundorf, M. K. 2003. Use of the Internet and e-mail for health care information:
results from a national survey. JAMA : the journal of the American Medical Association, 289, 2400-6.
Craigie, M., Loader, B., Burrows, R. & Muncer, S. 2002. Reliability of health information on the Internet: an
examination of experts' ratings. Journal of medical Internet research, 4, e2.
Cross, M. 2008. How the internet is changing health care. British Medical Journal.
Diaz, J. A., Griffith, R. A., Ng, J. J., Reinert, S. E., Friedmann, P. D. & Moulton, A. W. 2002. Patients' Use of the
Internet for Medical Information. Journal of General Internal Medicine, 17, 180-185.
Eysenbach, G. 2002. Empirical Studies Assessing the Quality of Health Information for Consumers on the World Wide
Web: A Systematic Review. The Journal of the American Medical Association, 287, 2691-2700.
Eysenbach, G. 2008. Medicine 2.0: Social Networking, Collaboration, Participation, Apomediation, and Openness |
Eysenbach | Journal of Medical Internet Research. Journal of Medical Internet Research, 10.
Eysenbach, G., Powell, J., Englesakis, M., Rizo, C. & Stern, A. 2004. Health related virtual communities and electronic
support groups: systematic review of the effects of online peer to peer interactions. British Medical Journal.
Griffiths, K. M., Calear, A. L. & Banfield, M. 2009. Systematic review on Internet Support Groups (ISGs) and
depression (1): Do ISGs reduce depressive symptoms? Journal of medical Internet research, 11, e40.
Hughes, B., Joshi, I. & Wareham, J. 2008. Health 2.0 and Medicine 2.0: Tensions and Controversies in the Field. Journal
of Medical Internet Research, 10.
Impicciatore, P., Pandolfini, C., Casella, N. & Bonati, C. 1997. Reliability of health information for the public on the
world wide web: systematic survey of advice on managing fever in children at home. British Medical Journal.
Kummervold, P. E., Chronaki, C. E., Lausen, B., Prokosch, H.-U., Rasmussen, J., Santana, S., Staniszewski, A. &
Wangberg, S. C. 2008. eHealth trends in Europe 2005-2007: a population-based survey. Journal of medical Internet
research, 10, e42.
Leimeister, J. M. & Krcmar, H. Year. Revisiting the virutal community Business Model. In: Tenth Americas Conference
on Information Systems, 2004 New York.
Mitchell, J. 2000. Increasing the cost-effectiveness of telemedicine by embracing e-health. Journal of telemedicine and
telecare, 16, 16-19.
Murray, E., Lo, B., Pollack, L. & Donelan, K. 2003. The Impact of Health Information on the Internet on Health Care and
the Physician-Patient Relationship: National U.S. Survey among 1.050 U.S. Physicians. Journal of Medical Internet
Research, 5.
O'reilly, T. 2007. What is Web 2.0: Design patterns and business models for the next generation of software. Social
Science Research Network Working Paper Series, 65, 17-37.
Wilkins, A. S. 1999. Expanding Internet access for health care consumers. Health Care Management Review, 24, 30-41.

ISBN: 978-972-8939-25-0 2010 IADIS
Ruth Diko Wario and Theo McDonald
Department of Computer Science
University of Free State
Previous studies on Internet activity in Kenya were mostly concerned with numbers: how many hosts, how many users or
subscribers and how much bandwidth. Little is known about what Kenya users actually do on the Internet, their usage
patterns, and interaction between Internet governing frameworks and policies, services and usage demographics and
behaviours. This study is intended to answer these questions and to lay a foundation for future studies in terms of access
technologies and behaviour patterns of the Kenya Internet population. It provides a basic understanding of the Kenya
Internet market to institutions involved in the development of the Internet market. It thus facilitates the development of
sound policies that will drive Kenyas information economy. As an outcome of the study a profile of the characteristics of
a typical Internet user in Kenya will be provided.
Internet access technologies, Internet usage patterns, Internet user profile, Kenya
The use of the Internet was first initiated by the military and academics. It has now, in a short while,
progressed from being a scientific means of communication to the main method of exchanging information
for almost everyone (Attaran and Vanlaar, 2000). As a growing and rapidly diffusing technology, the Internet
is an innovation that promises to become the economic reinforcement for all successful countries in the new
global economy (National Telecommunications and Information Administration, 2006). Understanding who
is connected to the Internet and how it is being used is critical to the development of sound policies in this
area (National Telecommunications and Information Administration, 2006). What is currently facing
governments and non-governmental organisations alike is the challenge to expand the connectivity of the
Internet in their different countries to even grassroots level.
Previous studies (Centre for Democracy and Technology, 2000; Oyelaran-Oyeyinka and Adeya, 2002),
however, revealed that several factors contribute heavily to this challenge. These include the poor state of the
underlying telecommunications infrastructure, high telephone call charges, high Internet Service Providers
(ISP) charges, advances in access technologies, policies and regulations around Internet market development,
and income and socio-economic status. The majority of African countries do not possess basic infrastructure.
In some cases, efforts have been made, but proved to be insufficient. In others, technological infrastructure is
obsolete as a result of the rapid developments of digital technologies (Oyelaran-Oyeyinka and Adeya, 2002).
The objective of this study was to investigate the Internet access technologies available in Kenya and the
Internet usage patterns among Kenyan users. The goal was to establish a profile of a typical user, his/her
perception of the Internet and what the future holds for the Internet in Kenya. This study is intended to lay a
foundation for future studies in terms of access technologies and behaviour patterns of the Kenya Internet
population. It provides a basic understanding of the Kenya Internet market to institutions involved in the
development of the Internet market. It thus facilitates the development of sound policies that will drive
Kenyas information economy.
The rest of the paper will be structured as follows. In the next section background information of how the
Internet developed in Kenya will be provided. That will be followed by the methodology used in conducting
the research. Consequently the results of the study will be provided and discussed.
IADIS International Conference WWW/Internet 2010
2.1 History of the Internet in Kenya
Mweu (2000) asserts that the Internet first became available in Kenya in 1993. As with many other African
states, Internet development in Kenya was primarily led by Kenyans returning from overseas studies,
Western expatriates and personnel of Inter-governmental Organizations (IGOs) and Non-Governmental
Organization (NGOs). Individuals in these groups had been exposed to the Internet and upon their return to
Kenya demanded Internet access (Centre for International Development and Conflict Management, 1998).
Initially, access to the Internet was achieved through a Gopher service, which was text-based. The
connectivity was made possible through either international leased lines or X.25 connections (Mweu, 2000).
The African Regional Centre for Computing (ARCC), an NGO based in Nairobi, became the first provider of
Web-based Internet services, based on the Mosaic browser. The connection to the global Internet backbone
was via an analogue leased line. The first commercial ISP, Form-net, began operating in 1995, and with the
entry of three other ISPs, competition soon increased. Today, there are over 50 ISPs operational in Kenya
offering different services (Communications Commission of Kenya, 2008a).
2.2 Regulation of the Internet Service in Kenya
The Kenya Communications Act KCA (No. 2 of 1998) provides the framework for regulating the
communications sector in Kenya (Waema, 2007). The Act was enacted by Parliament in 1998, to give
legislative teeth to the Postal and Telecommunications Sector Policy Statement, issued by the Ministry of
Transport and Communications in January 1997 and updated both in 1999 and 2001.
The Policy Statement defines the policy for the telecommunications operation and provides a framework
for the introduction of structural changes in the sector. The Policy Statement was set out against a deliberate
move by the government to optimise the sectors contribution to national development by ensuring the
availability of efficient, reliable and affordable communication services throughout the country.
The Act unbundled Kenya Post and Telecommunication into five separate entities including Telkom, the
fixed line operator; the Postal Corporation of Kenya (Postal); the Communications Commission of Kenya
(CCK) and the National Communications Secretariat (NCS). It also created an Appeals Tribunal for the
purpose of arbitration in cases where disputes arise between parties under the KCA. The national regulator,
CCK, issued a statement in September 2004 containing/introducing a new licensing framework. The general
goal of this framework was to ensure that the regulatory environment in the sector is friendly to investment
in and conducive to the provision of modern communication services - Internet (Waema, 2007).
The specific objectives of the new licensing framework were to ensure that Kenya has a more dynamic
and competitive Internet environment, improved access to Internet infrastructure and services and choice in
the provision of communication services to meet the socio-economic needs of the society.
2.3 Available Technology Choices for Internet Access
Access to the Internet service in Kenya is not widely spread as compared to other services e.g. mobile
phones, despite the liberalization of the sector (Communications Commission of Kenya, 2008b). This is
brought about by among other things, high costs of bandwidth. Kenyans have the option of accessing the
Internet using either modems with dial-up connections or leased lines. Dial-up connections are available
through both fixed telephone lines as well as Code Division Multiple Access (CDMA) technology introduced
in Kenya in late 2005.
Three operators provide CDMA wireless access technology: Telkom Kenya, Popote Wireless and
Flashcom. Other operators plan to offer a wireless option as a business strategy to remain competitive in the
increasingly liberalized telecommunications market. By some estimates 400,000 new Internet users will rely
on the wireless option in the first half of the year 2008 (Kinyanjui, 2009). Many Kenyans opt for CDMA as
equipment costs drop and coverage extends throughout the country. Currently, the coverage by Telkom is
limited to 50 kilometres of the major urban centres of Nairobi, Kisumu, Nakuru, Eldoret, Mombasa, and
ISBN: 978-972-8939-25-0 2010 IADIS
Internet access is available through either 64/128/256 Kbps leased lines or through 64 Kbps dial-up
accounts (Ochara, et al, 2008). Home users in Kenya mainly choose either the dial-up accounts option or the
wireless option because they are less expensive than leased lines. The leased line market is dominated by four
Internet Gateway and Backbone Operators (IBGOs). Data collected from the IBGOs indicate that they have
7,637 customers on leased lines while 17,737 customers are on dial-up accounts (Ochara, et al, 2008). This
proportion of users on dial-up accounts, as an indicator of customers using modems, represents
approximately 70 percent of the total Internet user population. While there are a number of corporate
customers on more than 64Kbs bandwidth, the larger proportion of customers have connections of 32Kbs and
Other technologies that can be used for access include XDSL, fixed wireless and Fibre access. At present,
only Telkom Kenya Limited (TKL) has a fibre optic installation. The fibre optic installation provides
junction circuits between digital exchanges replacing the old copper pair junction routes. Demand for
broadband services is increasing and the TKL plans to install fibre optical cable networks to customers
countrywide. The coming of high-speed fibre optic cables, and at least two more expected to go live by 2010
(BharatBook, 2009), will result in costs reduction and rise in connection speed, thus making technology
affordable and accessible to larger segments of the population.
The research design used in this study is a descriptive survey that seeks to explore Internet usage and its
motives. The methodology employed a combination of qualitative and quantitative approaches. The findings
were based on the results of questionnaires and interviews from the Kenya Internet service providers (ISPs),
communication commission of Kenya (CCK) and general users.
Several factors influenced the sample selection. For instance, only the people living in urban areas are
connected to electricity as compared to those living in rural areas. Therefore, users are highly concentrated in
Nairobi and are estimated to account for 80% of the countrys Internet subscriptions (Lilech, 2008). Nairobi,
the national capital, is also known for its cosmopolitan nature, and is the commercial hub of the region. It
consists of a large portion of the literate population, and is the main terminus of the regions communication
infrastructure, particularly the Internet. This city can, therefore, be seen as representing the country as a
whole in terms of Internet access and usage, and will be the main focus of this exploratory survey.
In order to obtain a truly random sample representative of Internet users in Nairobi, a complete list with
contact details of every Internet user was required. Such a list was not available and alternative methods had
to be followed. The following two methods of questionnaire distribution were, consequently, used: e-mail
distribution of questionnaires and usual handing out of printed questionnaires. The e-mailed questionnaires
were sent to individuals at homes, companies, schools, and government departments. Their email addresses
were extracted from the personal address books of several colleagues. The printed questionnaires were
delivered by hand to various Internet cafes for customers to complete on a voluntary basis. In this way 500
questionnaires were distributed. It is acknowledged that this was not a scientific way to obtain the sample,
but giving a lack of sufficient resources and no access to required information, it was most suitable at the
Out of 500 questionnaires that were distributed, only 148 were returned, constituting a response rate of
30%. Even though this return rate is low, it is similar to that conducted in other studies (Rivard, 2000).
Although the scenarios obtained from this survey may be sufficiently accurate to describe general levels of
Internet usage by the respondents, the low sample size and the way the sampling was conducted require
caution when using the data for strategic decision making and should, therefore, only be used as an indicative
guide for the industry in general.
The questionnaires served to gather information relating to the access and use of the Internet in Kenya. It
aimed to determine the penetration of the Internet and the characteristics of a typical user in Kenya by
investigating methods of accessing the Internet, reasons why people use the Internet, as well as how often
and from where they access the Internet. Two separate questionnaires were designed, one to collect
information from general users and one to collect data from the ISPs. Data collection took place during
December 2009.
IADIS International Conference WWW/Internet 2010
4.1 ISPs
A list of 50 registered ISPs with contact details were obtained from CCK and they were all contacted for
interviews. Twenty nine of them could not be reached even after visiting their physical addresses. Sixteen
were reached and questionnaires were sent, but even after several attempts to retrieve the questionnaires no
response was achieved. Two ISPs responded positively and provided the required information. Five other
ISPs responded, but were only willing to provide the date of establishment. Attempts to ensure
confidentiality of the information failed to convince these ISPs to provide the information required. The
information used in this analysis is, therefore, from the two ISPs who responded Wananchi online Ltd and
The consumer base is mostly made up of businesses and private consumers, and the market is mostly for
both email and browsing customers. The popularity of cyber-cafes is increasing, as signified by the large
number of users who subscribe to them. Other services provided by the ISPs include domain registration,
web design and hosting, web security, Virtual Private Networks (VPNs) and Virtual Satellite Aperture
Terminals (VSATs).
According to the report obtained from the Communications Commission of Kenya (2008a), the total
number of Internet users was 3.3 million, compared to 2.8 million of the previous year, representing a 9%
increase. The number of dial-up customers was higher (7,846) as compared to leased lines (1,809) and
wireless connectivity (5,226). There was a decline in the number of leased line customers and dial-up
customers by 6.7% and 1.9% in 2008 compared to 29% and 32% in 2007. The decline has been characterized
by the demand for wireless connections that has grown by 0.97%. The 2 ISPs reported that
and are the most visited sites respectively.
4.2 The Users
4.2.1 Demographic Information
With more and more people becoming Internet users, the demographic composition of the users are
continually changing. Currently in Kenya the proportion of males (64%) is higher than that of females (36%).
The average age of Internet users is one of the most significant indicators of Internet penetration in a country.
In Kenya the prevalent age group, which is approximately 57% of the total number, is between 16 - 25 years
as presented by Figure 1

Figure 1. The Age of Internet Users
In terms of educational level the results indicated that 94% of the respondents are people who have
studied either at university or college. Students represent the most numerous group, about 48% while
professions /professionals (working class) represent 45% of the Internet users.
The results further indicated that the higher an individuals income is, the more likely he/she is to have
access to the Internet. About 47% of adults with an income of more than Ksh10, 000 reported that they use

ISBN: 978-972-8939-25-0 2010 IADIS
the Internet. This is significantly higher compared to those who earn between Ksh5000 10,000, and those
earning below Ksh5000 as shown in Figure 2.

Figure 2. Monthly Income of Internet Users
Regarding PC ownership, 68% of the respondents have a PC at home, while 32% of the respondents did
not own a PC at all. For those who have PCs at home, 41% of the respondents could access the Internet at
home. This suggests that Internet services still remain out of reach for the majority of Kenyans as compared
to other countries, especially the USA where more than 80% of Americans have computers at home. Almost
92% of the Americans who own PCs have access to the Internet (Nielsen Online, 2009).
The frequency at which respondents access the Internet is shown in Figure 3. Among the respondents,
about 59% of the respondents use the Internet daily, 30% weekly and 7% monthly.

Figure 3. Internet Usage Frequency of Access
In terms of the question on purpose of Internet use, the results indicated that about 54% of the
respondents use the Internet for email purposes and 46% use it for World Wide Web applications such as
Facebook, news, Google and sports. The most popular types of information searched for are information
relating to education (37%), followed by culture/sport/entertainment (26%), news (16%) and
economic/finance (14%).
It is clear from the results that the opportunity of online shopping is not widely used. About 65% of the
respondents reported lack of interest in e-commerce or buying online. About 27% were interested in e-

IADIS International Conference WWW/Internet 2010
commerce while 8% were not sure. The common reasons given for not using e-commerce on the Internet
were security reasons, lack of credit cards and poor financial resources.
The general experience of those who have purchased goods online shows that they were satisfied with the
practice. Nearly 20% of the respondents reported the practice to be good, 33% reported that they were not
interested, 40% reported satisfaction, while 7% were not sure whether they would use it again. The most
popular search engine among Internet users in Kenya is Google (62%), followed by Yahoo (27%) and other
search engines (6%).
Generally, the respondents expressed the desire to use the Internet service more. However, the cost and
convenience of access has limited Internet use. Consequently, access times are short. Another factor is the
slow access speed. This is due to the poor quality of telecommunication links, partially attributed to
congestion. This frustrates many potential users and discourages Internet surfing. According to the
questionnaire survey, 52% of the respondents regarded slow page download time as their biggest problem,
37% reported cost to be their biggest problem, 5% failure to find the information on the web and 6% constant
error pages.
5.1 Internet Access Technologies Available in Kenya
In Kenya, Internet access is provided through dial-up or leased lines. Recently, broadband has been
introduced in the form of ADSL, fixed wireless, fibre access and cable modem. The government is currently
supporting several projects aimed at boosting the countrys broadband infrastructure with the most high-
profile projects being the fiber optic international submarine cable to the countrys shores, East Africa
Marine Systems (EAMS) and the East Africa Submarine Cable System (EASSYs). These initiatives will
connect the countries of Eastern Africa via a high bandwidth fiber optic cable system with the rest of the
world. TEAMS, a multi-million dollar fibre optic cable link from Mombasa (Kenya) to Fujaira in the United
Arab Emirates, is expected to link East Africa to the rest of the world (Kenya Business, 2008).
5.2 Profile of a Typical Kenyan Internet User
A user of the Internet in Kenya is typically a single male aged between 16 and 25 years of age with a
university/college degree and earning a monthly salary of less than Kshs 5000
. Although he owns a PC at
home, he is accessing the Internet from his workplace and using it on a daily basis. The most common uses of
the Internet include e-mail and searching for information on education, sport, news and entertainment. The
preferred search engines are Google and Yahoo among others. He is not yet ready for e-commerce and
Internet banking.
5.3 Internet Usage Pattern among Kenyan Users
This study is the first report on Internet usage patterns of the Internet population in Kenya. Although Internet
usage in Kenya is low and lags behind other telecommunication services, for example mobile phones
(Communications Commission of Kenya, 2008b), it still remains beyond the capability of many Kenyan
households. Quite a number of people have expressed a desire to use the Internet more, mainly because of its
economical, political, social and cultural development and possibilities.
From this research it appears that a considerable number of the Kenyan population gets access to the
Internet from their workplaces rather than any other source. The arrival of the fibre optic international
submarine cable to the country will lower the rates of international bandwidth which will finally increase the
number of Internet connections (Electronics Newsweekly, 2009).
The results of this study clearly indicate that the exploitation of the Internet for commercial transactions is
still in its infancy. This is due to the absence of enabling legislation (Information and Communication

US$1 is about Kshs 77
ISBN: 978-972-8939-25-0 2010 IADIS
Technology Kenya, 2005). As it is evident that there is potential for usage of e-commerce in Kenya, it is
interesting to note that some people still feel they are not yet ready for e-commerce or Internet banking. The
most popular reason is that the Internet is not secure for monetary transactions. Credit cards are also not so
popular in Kenya and this is another reason why there isnt much activity in e-commerce. A substantial
number of people still prefer physically going shopping.
5.4 The Future of the Internet in Kenya
For Internet uptake to be enhanced in the country there is a need for infrastructure improvement to improve
accessibility. Wider Internet connectivity can enhance Kenyas economic growth. Wider access to the
Internet will result in more people spending more time on the Internet. In future, many organizations are
likely to use the Internet to do business and create many jobs related to the technology. As the Internet
continues to expand tremendous potential exists for its use.
The future outlook of the Internet in Kenya is positive as commissions embark on the implementation of
the Kenya Communications Amendment Act of 2008. Telecommunication infrastructure will have a major
boost with the completion of the National Optic Fibre Backbone infrastructure. This key infrastructure is
expected to increase bandwidth capacity in most parts of the country. This should have a positive impact on
Internet diffusion in rural and remote areas. For example, the arrival of the first ever fibre optic international
submarine cable to the countrys shores in mid-2009, with at least two more expected to go live by 2010, will
lower the rates of international bandwidth which will finally take the Internet to the mass market.
The investigations of this study revealed that access technologies such as dial-up or leased lines are
commonly used while broadband technologies have only recently been introduced. The findings also showed
that in Kenya access and usage of the Internet are mainly associated with age, education, income and socio-
economic status, and that the market is occupied mainly by young people who are the frequent users of the
Internet and who would also like to try doing online business transactions in the near future.
Kenyas Internet market looks positive and the governments awareness of the advantages can increase its
usage. The much-anticipated undersea fibre optic cable and the improved competition regulatory framework
will boost Internet penetration. The introduction of broadband services by mobile operators is expected to
further boost Internet penetration and use.
This study has shown that, with the rate at which the number of Internet users is increasing, the Internet is
becoming more familiar to the Kenyan. This is especially true for the younger generation (26-35 years) that
regards it as a revolutionary technology that in the near future will permanently change the nature of data
collection, storage, processing, transmittal, and presentation. Information technologies have become essential
to enhancing the economic, social, and educational development of the country.
Attaran, M. and Vanlaar, I., 2000. Privacy and Security on the Internet: How to secure your personal information and
company data. Information Management and Computer Security, Vol.7, No.5, pp 241 - 246.
BharatBook, 2009. Kenya Convergence Broadband and Internet Markets. URL
Research-Reports/Kenya-Convergence-Broadband-and-Internet-Markets.html [Accessed 14th October 2009]
Centre for Democracy and Technology, 2000. The Digital Divide: Internet Access in Central and Eastern Europe. URL [Accessed 30th November 2009]
Centre for International Development and Conflict Management (CIDCM), 1998. The Internet in Kenya: Impacts and
Developing. URL [Accessed 12th June 2009]
Communications Commission of Kenya (CCK), 2008a. Communication Statistic Reports Second Quarter 2008/2009.
URL [Accessed 6th November 2009]
Communications Commission of Kenya (CCK), 2008b. Annual Reports 2007/2008. URL [Accessed 24th August 2009]
IADIS International Conference WWW/Internet 2010
Electronic Newsweekly(2009). Kenya-Convergence, Broadband & Internet Markets. URL
http://www.highbeam. com/doc/1G1-210810131.html [Accessed 2nd October 2009]
Information and Communication Technology Kenya (ICT Kenya). (2005). Kenya's Information & Communications
technology Sector. URL [Accessed 5th June 2009]
Kenya Business. (2008). Telecommunications and IT Overview. URL http://www.kenyarep- [Accessed 2nd August 2009]
Kinyanjui, K., 2009. Kenya: ISPs warn against Price Regulation. URL
[Accessed 6th August 2009]
Lilech, H., 2008. Kenya: Wireless Industry. URL
[Accessed 30th November 2009]
Mweu, F., 2000. Overview of the Internet in Kenya. URL [Accessed 26th November 2009]
National Telecommunications and Information Administration, US Department of Commerce, 2006. Falling Through
the Net: Part II Internet Access and Usage. URL
[Accessed 30th November 2009]
Nielsen Online, 2009. Home Internet Access: Continuing to Grow, But Big Differences Among Demographics. URL
among-demographics/ [Accessed 5th September 2009]
Ochara, N. et al, 2008. Global Diffusion of the Internet X111: Internet Diffusion in Kenya and Its Determinants - A
Longitudinal Analysis. Communications of the Association for Information Systems, Vol.23, No.7, pp 123 - 150.
Oyelaran-Oyeyinka, B., Adeya, C. (2002). Internet access in Africa: Empirical evidence from Kenya and Nigeria. URL
&_userid=736898&md5=1cc508253400b7b8ffddcfff3c134dba [Accessed 2nd July 2009]
Rivard, H., 2000. A survey on the impact of information Technology on the Canadian Architecture, Engineering and
Construction Industry. Electronic Journal of IT in Construction, Vol. 5, pp 37.
Waema, T., 2007. Kenya Telecommunications Sector Performance Review. URL [Accessed 25th August 2009]
ISBN: 978-972-8939-25-0 2010 IADIS
Malek Barhoush and J . William Atwood
Department of Computer Science and Software Engineering, Concordia University
1455 de Maisonneuve Boulevard, West, Montreal, Quebec, H3G 1M8
Software-based protection for digital content is a promising and flexible solution for the content provider whose
distribution is to a community in which the majority of clients are computer and laptop users. In this paper, we explore an
example of software-based copy protection for temporal media distribution that has been published by Grimen, et al. We
evaluated the security protocols provided in the example and found two security attacks. We provide the solution to
overcome the discovered attacks.
DRM, Tamper Resistant Software, Formal Validation.
Watching movies and TV channels through the Internet gives the client the flexibility to explore for
interesting programming, and then start viewing. The content provider (CP) on the other side has the
motivation to use the Internet to deliver the content media because the cost of distributing the information is
low. However, the problem is that the Internet does not facilitate access control for electronic goods.
Digital rights management (DRM) is an application framework introduced to help a CP to control the
access to electronic products and to enforce the usage according to the permissions and constraints stated by
the CP or the owner of the electronic product. The central component of a DRM system is a tamper resistant
entity; this entity may be implemented in hardware or software. The whole security protection and
enforcement mechanism of a DRM system depends on the fact that the tamper resistant element is guaranteed
to be unbreakable [Stamp_digital].
Tamperproof hardware provides physical protection and tries to hide the protection technologies used to
secure the content media. Because the end user may have hacking tools and the time to compromise
tamperproof hardware, hiding protection technologies does not provide a permanent solution. It is typically
more effective than a software solution, but it is harder to recover when the hardware is compromised. A
dongle provides hardware copy protection and should be connected on one of the computer ports in order to
use dongle-based software. Dongle-based software checks the presence of a token issued by the dongle. If
the software discovers that the dongle is not connected then it does not give the necessary signal to execute or
to run the full version of the software. The security of this methodology depends on the difficulty of cloning
dongles. The protection provided by a dongle is susceptible to reverse engineering and code modification
The Open Mobile Alliance DRM2.1 (OMA-DRM2.1) is a framework solution to control the access to
mobile ringtones, pictures or songs. OMA-DRM2.1 enabled devices have tamper resistant hardware that
contains a unique id for identification purposes and a unique private/public key pair as well as a certificate.
This feature allows the operator or the rights issuer to easily authenticate the OMA-DRM2.1 enabled devices.
Each right object issued to a specific device is cryptographically bound to that device. The whole security of
OMA depends on hiding the private key [Ntzel-DRM, OMA-DRM2.1]. Hardware tamper resistant devices
are difficult to deploy, especially when the majority of viewers is using a personal computer or a laptop. If
the solution provided by tamper resistant hardware is compromised, then replacing the failed hardware is not
IADIS International Conference WWW/Internet 2010
simple. A fast solution is to revoke the compromised device [Grimen-DRM1], but this may not be very
satisfying to the end user.
The software-based protection task touches on many aspects: intellectual property protection,
preventing a software hacker from extracting sensitive information via reverse engineering or static and
behavior analysis, and controlling the use of digital media. All protection mechanisms provide short-term
protection [Maa-Framework].
Apple's DRM solution is built on a software-based tamper resistant approach; it depends on hiding a
repository key in a secret algorithm embedded in the iTunes application. This repository key is needed to
decrypt a user key stored in the user side, the user key is needed to decrypt a master key, which is needed to
decrypt the corresponding protected media file. The whole idea of the protection mechanism depends on
making this secret algorithm hard to expose. This called security via obscurity. This method did not survive
because an attacker was able to reverse engineer the iTunes application and extract hidden secrets such as the
secret algorithm. For more details see [Grimen-DRM1, WWW-FairPlay].
Microsoft DRM is another example of a software-based controller used to control protected media. It
depends on keeping a blackbox file hidden from the end user. This blackbox contains the client key, which is
generated in the individualization of the user's blackbox. This key is stored in a scrambled form and
theoretically is hard to extract. This key used to cryptographically bind the license file with the users DRM-
enabled application. The license contains the content key that is used to encrypt the content media. This
solution also did not survive, because it depends on keeping the secret hidden from the end user. The end
user has the ability, with sufficient time and tools such as memory dumping and reverse engineering, to
extract this information, for more details see [Grimen-DRM1, Req-DRM].
A DRM system works in the customer environment. Thus the customer has full control to analyze and
reverse engineer the DRM workflow. One way to make the attackers work more difficult is to integrate
hardware tamper resistance with DRM workflow, such hardware is used to embed some protection
components. An example of hardware tamper resistance is using smart cards for authentication purposes or
using set-top-boxes for protecting electronic media. The main problem with hardware tamper resistance is
that it is expensive to integrate with the majority of end-user laptops and personal computers, which are not
prepared for DRM purposes. Another way is to integrate software tamper resistance with a renewability
mechanism. Renewability is a mechanism used to periodically modify the software protection component. It
is hoped that this will be done before the protection component has been compromised. Software-based
protection is flexible, inexpensive to transfer within a public network and works on different platforms
[Grimen-DRM, Grimen-DRM1, Req-DRM 1].
Bar-El claims that no security technology can predict or prevent future attacks. Renewing protection
technology prevents an attacker from taking advantage of the defects discovered in current protection
technology. Bar-El suggests that for a renewable protection scheme, the scheme at least needs to be divided
into two pieces: a permanent component such as the viewer application and a modifiable component such as
plug-in components [Bar-El]. Renewability in the context of the Motion Picture Experts Group (MPEG-4)
Intellectual Property Management and Protection (IPMP) extension is accomplished by renewing
compromised tools that are used for authentication, decryption and watermarking of content media [MPEG-
4_IPMP, MPEG-4_IPMP2, MPEG_Eval]. As far as we have been able to discover, MPEG does not prevent
a reverse engineering attack. J onker et al. suggests that all security systems need to adapt protection system
measures to compensate for any attack [Jonker-CoreSec]. Grimen et al. suggest that changing the viewer
software that is responsible for rendering the content media is equivalent to reconfiguring that viewer, this
kind of reconfiguring of viewer software may be able to detect and prevent any hacker attacks [Grimen-
Grimen et al. [Grimen-DRM2] proposed a software solution based on periodic renewal of the protection
scheme that is used to control the access to content media. The software-based solution they proposed
appeared to be a promising solution, but we have found two attacks against the protocol that they use to
control the access to protected media. We will discuss their solution and the attacks in section 2. In Section
3, we will give a formal model of the attack. In Section 4, we give a solution to the attack, and demonstrate
its security. In Section 5, we offer our conclusions.

ISBN: 978-972-8939-25-0 2010 IADIS
Grimen et al. [Grimen-DRM2] proposed an architecture for software-based secure content distribution that
consists of four players: the content provider (CP), the stream server, which we are going to call the content
server (CS), the viewer software (VS) and the security server (SS). The CP encodes the content media, and
divides it into many pieces. It then produces as many symmetric keys as there are media pieces, in order to
protect each piece with one unique symmetric key. This process results in what they called an encrypted
encoded media document (EEMD) and makes it ready for distribution. See Figure 1. The authors called the
time period for each piece a trust interval.
The CS is in charge of distributing the EEMD piece by piece. The VS player is running within the user
environment. It is responsible for decrypting and decoding the EEMD and then rendering the content media
and making it available for viewing. The SS is responsible for generating and delivering a piece of code
called a Mobile Guard (MG), which is able to execute in the user environment. The MG needs to be plugged
into the VS. The MG needs to be sent to the VS in order to configure the VS to maintain security states of
the media document. The MG is responsible for checking the integrity of the VS components, including the
MG itself, and then after successful integrity checking, the SS is going to deliver a corresponding media key
for a current delivered piece of the EEMD. The authors claim that the MG is hard to compromise; if
successful software modifications happens, it will be detected and the system will have break once break
everywhere resistance [Grimen-DRM2].

Figure 1. Dividing the media content [Grimen-DRM2].
As the encoded media document is being split into multiple pieces, each piece is encrypted with a
different key. The SS is going to provide the decryption key for each piece only at the start of its particular
trust interval, upon request and after checking the integrity of VS; the VS needs to prove its good intention
before delivery of the new decryption key. This good intention is achieved when the VS receives and
executes a new generated MG, which means that for every decryption key there is a new MG. Each MG is
distinct from the others, and each one is responsible for reconfiguring and securing the VS. The MG needs to
be obfuscated to prevent the software hacker from predicting the internal logic of the MG [Grimen-DRM2].
The MG is replaced for each trust interval, in order to prevent the software attacker from predicting the
VS configuration when he has the time to gain the knowledge. Each instance of the MG is responsible for
protecting the access control mechanism used by the VS for the duration of the trust interval. Each MG
instance is going to check the integrity of the VS every new trust interval. The VS is going to request the
corresponding decryption key at the end of each trust interval. The SS is responsible for receiving the
checksum for the VS for each trust interval. Then, upon verifying the correctness of each checksum, it is
going to send the next decryption key that corresponds to the new media piece. Figure 2 shows the Grimen
et al. proposed architecture. The details of the process of receiving the decryption key for each trust interval
(as proposed by Grimen et al. [Grimen-DRM2]) are as follows:
The VS generates a random unpredictable transport key.
The MG executes and determines the integrity checksum of the VS along with the MG itself.
The integrity checksum is determined by computing a one-way hash function across the VS
instructions, the MG instructions, MGs public key and the generated transport key. See Figure 5.
The VS encrypts the transport key with the SS public key.
The VS sends the encrypted transport key to the SS along with the integrity checksum.
The SS verifies the checksum. Since the SS can extract the transport key, and since it knows the VS
instructions, the MG instructions, MGs public key and the generated transport key, the SS can then generate
IADIS International Conference WWW/Internet 2010
the checksum. The SS verifies the calculated checksum against the received checksum. If the verification is
successful, the SS encrypts the corresponding trust interval's media key with the transport key and sends it to
the VS.
The VS now can decrypt the corresponding piece of the EEMD. See Figure 3.

Figure 2. Grimen et al. proposed systemarchitecture [Grimen-DRM2].

Figure 3. Key exchange protocol [Grimen-DRM2].
We analyzed the presented protocol and found two security attacks. The first attack happens because
there is nothing that can prevent the software attacker from forwarding the MG to another VS, and then the
forwarded MG produces a new transport key and sends that key to the SS. There is nothing to tell the SS that
the MG is a copy, not the original. In this case the server will respond with the media key encrypted with the
new generated transport key provided by the pirated MG. Another attack appears because the generation of
the transport key is done by the VS, because the VS executable file is stored in the client environment. Thus
the client has the chance to statically or dynamically analyze the code and learn the memory address of the
transport key.
The main problem with this key exchange protocol comes from the fact that there is nothing to distinguish
any instances of MG. All of them have the same features and no individualization technique has been
attached to them. The secondary problem is due to the fact that the client has enough time to statically or
dynamically analyze the VS. One way to prevent the previous attacks is to individualize the MG for each
client and to store the transport key in an unpredictable place. In the next Section we will discuss how to
achieve both goals.
Automated Validation of Internet Security Protocols and Applications (AVISPA) is a tool used for formal
modeling and analyzing of Internet security protocols. The model is specified in high level protocol
specification language (HLPSL), which represents the message exchanged between participants. Those
participants represent roles that send messages and receive messages, these messages may contain the
necessary keys needed to establish secure sessions and convey authentication actions between roles
[AVISPA-Manual]. We translated the Grimen et al. key exchange proposal into the following messages:
1- MG ->S: {Nm|Tki|MG}_Ks | hash(MG|VS|Nm|Tki)
Where MG: The identity of the mobile guard
S: The identity of the security server
Nm: nonce generated by MG for authentication purposes.
ISBN: 978-972-8939-25-0 2010 IADIS
Tki: transport key for session i, a generated symmetric key randomly generated by MG, used to transport
the media key for the ith trust interval session.
Hash: one way hash function.
Ks: public key for S.
VS: is the instructions of the viewer software along with MGs instructions, see Figure 5.
|: concatenation.
2- S->MG: {Nm|Ns|S}_Km
Where Km: public key for MG.
3- MG->S: {Ns|MG}_Ks
4- S->MG: {Mki|S}_Tki
Where Mki: media key for the trust interval ith session.
Figure 4 depicts the protocol exchanged between the two roles.

Figure 4. Protocol Simulation. Figure 5. Input of hash function calculation [Grimen-DRM2].

Figure 6. Attack trace for key exchange protocol proposed by Grimen et al.
We translated the previous message exchanges into HLPSL and simulated the attack. Figure 6 shows the
attack simulation. i represents the intruder entity who is trying to gain access to the media key for all
sessions. An intruder is a forwarded MG, i.e., an illegal copy of MG, who knows MGs public key.
The first message shows that the intruder generates a transport key and a new nonce, and encrypts them
with SSs public key. Then, the intruder generates a checksum that is a result of applying code instructions
into a hash function, these instructions are: fixed parts of the binary code of the VS and MG, transport key
and nonce. Then the intruder sends the result to SS. SS extracts the transport key and nonce, and then
calculates the checksum code, since the VS has all inputs for the hash function, and then compare the result
with received information. Upon successful verification, the SS creates a nonce and then sends the SSs
nonce along with MGs nonce all encrypted with MGs public key. The intruder who has the MGs public
key can extract the SSs nonce and then encrypt it with SSs public key and then send it to SS. The SS
believes that he is talking to a legal MG, and then encrypts the media key for ith session with the transport
key. This leads to an attack since the intruder can extract that media key. We assume that forwarding the
Mobile Guard to an illegal user is a DRM attack, which means that the illegal user uses indirectly the MG
public key or at least the pirated MG can decrypt any messages that have been encrypted with MG's public
The problem with the Grimen et al. solution is that generating a new transport key from any MG instance
does not correspond to the validity of any MG instance, thus any message from a pirated MG is accepted. A
IADIS International Conference WWW/Internet 2010
solution to prevent the previous attack is to give the ability for the SS to distinguish each MG instance. We
can achieve this distinction by individualizing all legal MG copies.
Apple's DRM solution used three keys to protect its products: master key, repository key and client key. The
product media is encrypted with the master key, the master key is encrypted with the client key, and the
encrypted master key is attached to the protected content. When the user asks for a license to use the product
media, the server that locally stores the client key sends the requesting user a license that includes the client
key encrypted with the repository key. The application that presents the product media at the client side has
the repository key and is attached with a secret algorithm embedded in the presentation software. The
protection mechanism is dependent on keeping the secret algorithm hidden from the user. Each user has a
distinct client key and repository, which simplifies the license individualization process. The client key is
randomly created by the server and saved in the server's database for further use. The repository key is
calculated from serial number of the first hard disk, BIOS version, CPU name and windows ID, which is
assumed to be unique [Grimen-DRM1].
In the Microsoft DRM solution the product media are encrypted with a content key and the content key is
encrypted with client keys. When the user asks for a license to use the product media, the license server sends
the license that contains the encrypted content key to the user. In the client side, there is a blackbox file that
contains the client key. Each client has an individualized blackbox, which means each instance of a blackbox
can only work for a specific machine. The protection mechanism relies on keeping the blackbox hidden from
the user access, which simplifies license individualization process [Grimen-DRM1].
OMA DRM standard encrypts the product media with a content encryption key (CEK), the CEK is
encrypted with a DRM agents public key and then inserted into a right object (RO). This process
cryptography binds an RO into DRM agent. Each DRM agent has a unique public and private user key; when
a DRM agent asks for right object to use a specific product, the server, who is responsible for creating right
objects, can obtain the DRM agent's public key, and create a unique right object for each DA that includes
the CEK key, which is encrypted with a DAs public key. Only the DA who has the corresponding private
key can use that RO and get the CEK [OMA-DRM2.1].
From the previous solutions we have two software solutions and one hardware solution for the end-user
individualization process. The case that we are working with needs to prevent an end user from forwarding
the MG to another user. To do this, we need to individualize each MG and make MG code not to work on
any other user. We suggest that the SS, which is responsible for generating an MG, needs to authenticate
each user and then embed a unique identity (Ticket) in the MG. The security server is going to accept only
one request for media decryption key from each legal MG instance per trust interval. Due to the fact that the
generation of transport key is unpredictable to both SS and user, the ticket value along with generated
transport key will be the unique identifier for each MG per trust interval. When the SS receives two requests
for the same MG that have the same ticket and a different transport key, the second key request will not
receive any response. This will prevent a pirated MG instance from successfully attacking the business model
previously discussed. Here is the modified protocol:
1- MG ->S: _{Nm|Tki|MG|Ticket}_Ks}_Km | hash(MG|VS|Nm|Tki|Ticket)
2- S->MG: {Nm|Ns|S}_Km
3- MG->S: {Ns|MG}_Ks
4- S->MG: {Mki|S}_Tki
Where Km here is a symmetric shared key between the MG and the SS, and the Ticket is the unique value
for each generated MG.
Before SS delivers an MG to a valid VS, it embeds a symmetric key Km into the generated MG. If a
legitimate user receives a valid MG, and then forwards that received MG to his friend, both MGs will have
the same identity (ticket). Each MG is going to run and generate a unique transport key at the client host; this
makes a checksum calculation unique since the transport key is one of the hash function inputs. We will call
the result of the hash function a checksum. When both MGs send the first message, which contains: nonce,
random generated transport key, mobile guard identity and ticket all encrypted with the SSs public key and
then again encrypted with shared key (Ks), the second part of the massage is the checksum. The SS can
extract the ticket value and the transport key and use them to calculate the checksum. The SS compares the
ISBN: 978-972-8939-25-0 2010 IADIS
received checksum with the one it calculates, if both match then the MG has sent correct information. The SS
checks the ticket value, if it receives it for the first time, then the MG is legal, so it will respond with the
second message. If it receives the ticket for the second time, and if the checksum is the same, then it can
safely reply to the second message, assuming that its previous response was lost in the network. If the SS
receives the same ticket with a different checksum, this means that illegal MG sends the second request, in
this case the SS will not reply to the request. The second message contains a generated nonce from the SS
side, MGs nonce and the SS identity all encrypted with the shared key Km. The third message is from the
MG side and contains: SSs nonce and MGs identity all encrypted with SSs public key. Now the SS and
MG are mutually authenticated, the forth message is from SS side and contains the media key for the ith
session and the SSs identity all encrypted with the transport key for ith session.
To prevent the software hacker from discovering the transport key in the VS space by using static or
dynamic analysis, the MG needs to create a random Transport Key and store it in a random place in the VS,
or keep it in MG space; the VS should implement a way to call the transport key generation from the plug-in
MG instance. This will prevent the end user from knowing the location of the transport key for a short time.

Figure 7. Message exchange for revised protocol.
In the solution we provide, the SS only accepts the first media key request for each unique ticket, and
rejects any subsequent request for the same ticket with different checksum. Figure 7 shows the protocol
simulation. We ran the new protocol on AVISPA and did not find any attack. We therefore believe that our
protocol is correct and helps the SS to authenticate valid instances of MG.
Software based protection is a promising solution for CP to deploy especially when the clients are general
purpose computers or laptops. We studied the software based solution introduced by Grimen et al. to protect
any content media at a client machine. We found two attacks, one in the key exchange protocol and the other
in their architecture design. We used AVISPA to simulate the attack for first flow. We proposed a modified
protocol that removes the first attack, and changed a little bit in the architecture design to remove the second
attack. We demonstrated its correctness using AVISPA.
Malek Barhoush acknowledges the support of Concordia University and of Yarmouk University.
J . William Atwood acknowledges the support of the Natural Sciences and Engineering Research Council
of Canada, through its Discovery Grants program.

IADIS International Conference WWW/Internet 2010
[AVISPA-Manual] 2006. Automated Validation of Internet Security Protocols and Applications-AVISPA v1.1 User
Manual Document Version: 1.1.
[Bar-El] Hagai Bar-El, 2005. Challenges of Standardizing Renewable Broadcast Security, White paper,
[Grimen-DRM1] Gisle Grimen et al., 2006. Building Secure Software-based DRM systems, NIK-2006 conference,
[Grimen-DRM2] Gisle Grimen et al., 2005, Software-based copy protection for temporal media during dissemination and
playback. In the 8
international conference on Information Security and Cryptology - ICISC 2005. Springer-Verlag,
Berlin Heidelberg, volume 3935 of LCNS, pp 62-377.
[J onker-CoreSec] H.L. J onker and S. Mauw, 2007. Core Security Requirements of DRM Systems, Digital Rights
Management -- An Introduction. ICFAI University Press, pp 73-90.
[Maa-Framework] Antonio Maa et al., 2004. A framework for secure execution of software. In International Journal
of Information Security, Springer Berlin / Heidelberg. pp 99-112.
[MPEG-4_IPMP] Ming J i, SM Shen and Wenjun Zeng, 2004. MPEG-4 IPMP Extension- For Interoperable Protection of
Multimedia Content, EURASIP Journal on Applied Signal Processing, Special Issue on Multimedia Security and
Rights Management, pp. 2201-2213.
[MPEG-4_IPMP-2] MPEG-4 IPMP Extension, Frequently Asked question,
[MPEG_Eval] HweeHwa Pang and Yongdong Wu, 2005. Evaluation of MPEG-4 IPMP extension, Acoustics, Speech,
and Signal Processing,. Proceedings. (ICASSP '05). IEEE International Conference, vol.2, no., pp. 18-23.
[Ntzel-DRM] J rgen Ntzel and Anja Beyer, 2006. How to Increase the Security of Digital Rights Management
Systems Without Affecting Consumer's Security, ETRICS 2006, Springer-Verlag Berlin Heidelberg, LNCS 3995, pp.
368 - 380.
[OMA-DRM2.1] OMA, 2008. DRM Architecture Approved Version 2.1 14 Oct 2008, pp 1-28.
[Req-DRM] Barhoush, Malek and Atwood, J, 2010. Requirements for enforcing digital rights management in multicast
content distribution, Telecommunication Systems, Springer Netherlands, volume 45, pp 3-20.
[Stamp_digital] Mark Stamp, 2003, Stamp: Digital Rights Management: The Technology Behind The Hype, Journal of
Electronic Commerce Research, VOL. 4, NO. 3, pp 102-112.
[WWW-FairPlay] sdebswrbh, FairPlay, In Art & Entertainment,, Accessed Sep 24,
ISBN: 978-972-8939-25-0 2010 IADIS
Gregory Triantafillidis and John Garofalakis
University of Patras, Computer Engineering and Informatics Department
26500, Patras, Greece
Hotlink assignment is a constantly growing field of research addressing the problem of low information access rates in
the World Wide Web. In this paper we attempt a more practical approach to the problem, by proposing an administrative
tool, the HotLink Visualizer, which implements hotlink additions to stored instances of the web site in question, in a
user friendly way. Our tool obtains and stores all the necessary information concerning the web sites connectivity using
a modified and specially configured web crawler. The outcome is a series of stored web site instances, produced by
applying different sets of hotlinks (generated by specific hotlink assignment algorithms) to the initially stored instance,
which can be edited and visualized by the user. HotLink Visualizer aims to bridge the gap between the several
theoretical hotlink assignment methods proposed by researchers and the need to put their results into use.
Hotlinks Assignment, Software Tools, Algorithms.
The World Wide Web has become established as the most popular source of information retrieval. As
expected, the older it gets the more information it contains and thus the number of the web sites with gigantic
growth and bad information access rates are constantly increased within it. The efforts to optimize these
access rates are continuous and towards those, several fields of research have developed, such as the
improvement of web design, clustering and caching. Obviously, although a web site is only a cell of the
World Wide Web, a well designed and organized site contributes to the improvement of the information
access rates in the web. Good structure for a web site means less traffic on the Internet, as the user gets the
piece of information required without wandering unnecessarily in it. Furthermore, a properly organized site
constitutes a more appealing choice for the users.
During the last years the matter is being addressed with the development of several hotlink assignment
algorithms for web sites. The idea behind those algorithms is to spot the most popular or more likely to be
accessed data and provide better access to it by assigning additional links (hotlinks) pointing to the web
pages containing it. These algorithms are not applied to the actual representations of these web sites but
usually to their corresponding direct acyclic graphs (DAGs) or to other even more strict structures. However,
it is widely known that a web site in its true form is not a DAG, since there can be found many links pointing
to just one page, thus forming circles and repeated nodes within the graph. Hence, there is a gap between the
theoretical determination of a set of hotlinks and the actual application of this set to a real web site.
In this paper we first address the issue of acquiring and storing the exact map of a web site with its full
connectivity, which can be considered as a first step towards the assignment of hotlinks in real web sites. By
storing we mean keeping an offline, viewable and -for our purposes- editable instance of the web sites
map. To do so, we first clarify theoretically what a web sites map with the complete connectivity
information should be and we determine the subsequent prerequisites that our crawler must have. We then
proceed with the appropriate specification and modification of an existing web crawler, with functionality
suited to our specific needs. Finally we propose an administrative tool, the Hotlink Visualizer, which, after
storing in tabular data all the necessary information to capture a web sites real map, it visualizes the outcome
and implements hotlink additions by adding with an automated procedure the generated hotlinks in the web
IADIS International Conference WWW/Internet 2010
pages of the site. Thus we have the ability to maintain in row data, different forms and versions of the
originally parsed web site, as they can be formed from the assignment of different hotlink sets to the original
site. Our tool aims at being user friendly from a web masters point of view and oriented towards the easy
optimization of the web sites information access rates.
Related work in this field of research includes algorithmic methods for hotlink assignment and tools that aim
to improve the design and the information access rate of a web site. Also all relative research that is oriented
towards a practical implementation of its results.
Garofalakis, Kapos and Mourloukos put forward an interesting proposal in [1], where an algorithm for
reorganizing the link structure of a web site was presented. The algorithm was based on data extracted from
the log files. The main criterion was the page popularity extracted by the pages depth and hit count. The
study assumed a binary tree structure for the site. However, in spite of the strict representation, there was
consideration for reentering the links initially removed in order to achieve the binary tree structure, thus
minimizing the loss of information of the sites link structure.
Miguel Vargas Martin in [5], presented metrics, detailed methods and algorithms targeting the Hotlink
Assignment problem. He targeted different aspects of the problem and provided measurements solidifying his
results. He also provided the NP-completeness proof of the Hotlink Assignment problem.
In [9], D. Antoniou et al, presented the hotlink assignment problem in a context-similarity based manner.
They proposed a direct acyclic graph model for the studied site, new metrics and an algorithm for the
process. Their context similarity approach takes interest in avoiding misplacement of the hotlinks.
In the field of web site optimization tools, Garofalakis, Giannakoudi and Sakkopoulos presented some
work in [3]. They noticed that there are no integrated tools presented for web site semantic log analysis that
could be delivered as end-user applications to help the web site administrator, so they proposed and
implemented a new information acquisition system that aims to enhance and ease log analysis by use of
semantic knowledge. Their tool takes into consideration both the site content semantics and the web site page
visits. It also extracts information about the user preferences and in general can be considered as a valuable
application for the administrators decision-making about the web site reorganization.
However, apart from the work mentioned above, in general it is fair to say that researchers have put more
effort in specifying and theoretically proving their web site optimization proposals, as opposed to putting
them more into use and making them accessible to the end user.
3.1 The Notion of a Link Oriented Crawler
Crawlers are software that traverse the Web automatically, access and download pages mainly for use in
search engines. In general, a web crawler takes as input a list of URLs. Then it downloads the web content
that the URLs point to and by discovering new URLs the procedure continues retroactively [7]. In our case
we implemented a web site crawler. The obvious difference is that a web site crawler takes as an input one
URL, the home page. During the crawling procedure, every new URL discovered is compared to that base
URL and the crawler adds it to its queue only if the base URL is part of the new URL. Apart from this
difference, a web site crawler doesnt need to be built around the need to download and process huge amount
of data, or keep the gathered content fresh, by revisiting the web sites. Such issues concern web crawlers that
are built around the notion of the node - web page and in our research are not taken into consideration.
Our main goal is to acquire all the information of the sites link structure. Thus we are not concerned with
the content of each web page. In other words we want to obtain the real map of the web site and store it in a
suitable formation, in order to perform hotlink additions on it. This provides us with a totally different
context in terms of choosing a crawler to work with and configuring its algorithm and overall operation.
Firstly, we need to clarify what we mean by real map. The first step in every research proposal for
hotlink assignment is to come up with a model for the web sites representation. According to the
ISBN: 978-972-8939-25-0 2010 IADIS
researchers goals, a web site is usually represented as a full binary tree, a direct acyclic graph, or as some
other restrictive structure. It is common ground that most of the link structure information is lost in that first
step and whats more, it is never retrieved. So to say we have extracted and stored the real map of a web
site, it means we make sure that no possible link will be passed over by the crawling process and all possible
pairs of {father node, child node} will be recorded.
With this last concession alone, it becomes transparent that we build the algorithm of our crawler around
the notion of the link between two nodes web pages of the site, contrary to what applies to most -web page
oriented- crawlers. Or as we said before, we implement a link oriented, web site crawler.
3.2 The Notion of a Link Oriented Crawler
Our specific crawling needs as described above, infer several other matters that we need to consider in
specifying the crawling algorithm. Firstly, it is common ground that in actual web sites, a user can be lead to
a web page from more than one parent pages. It is important however for our study to be able to record the
distance of each web page from the root, because this information will provide a measure of the access cost
of the page, which can be exploited later on. In particular we take special interest in identifying the first
appearance of a web page, which is the page with the minimum depth from the root. Recording this
information means that our crawling algorithm must be able to identify that first appearance, process it in a
different manner and of course store that information. To succeed in that, we request that our crawler follows
breadth first search. The reason is better depicted in the two following figures. In the first we see how the
web site would be traversed by a depth first algorithm and in the second by a breadth first algorithm.

Figure 1. a) Depth first. The numbers next to the nodes depict the order in which the crawler visits the nodes. In this tree-
like representation we see that each page occurs more than once, since in real sites each page can emerge from many
different parents. With depth first search, Node E would first occur at a depth of d=2.
b) Breadth first. This time node E is rightly recorded as having its first appearance with depth: d=1.
In the case of recording the web sites real map, the mistake described above would be significant. For
instance, if we miscalculate the depth of the first appearance of a web page, this would result in the
miscalculation of all depths of its children pages and so on. In the end we would not be able to regenerate the
correct map from the stored data. We could seek the first appearance of a web page by just looking for the
minimum depth (since the depth is retrieved anyway). However we want to store all the appearances of a web
page in the site and also store them in the order that they actually occur. The downloading and parsing of the
page is performed only once and only in its first appearance (the following appearances are simply recorded).
So the correct determination of that first appearance is a necessary step for the correct mapping of the site.
Since we need to know all the transitions between the nodes of the site, we must record all possible
parents that a child node can occur from. This means that crawling only once for each page-node is not
enough. When a common crawler opens a new page, it looks for links. If those links have already been
found, or lead to pages that have already been crawled, they are ignored since they are treated as old pages
that have no new information to provide. In our case however we record the link structure and are not
interested in storing the content of the pages. Consequently, our crawler does not ignore links that lead to
pages that have already been traversed. Instead we check if those links occur for the first time, considering of
course the current father node. In other words, if the current pair {father_node, child_node} occurs for the
first time, this to us is new information and therefore needs to be recorded. For instance, consider the case
that while the crawler parses a page A, finds a link to a page B that has not appeared before. Page B, will be
IADIS International Conference WWW/Internet 2010
added to the queue for future expansion. Later when the crawler parses another page C, finds again a link to
page B. We dont ignore this link. We record its appearance as new information for the sites link structure.
However we do not send again page B to the queue for downloading and expansion. The following table
sums up how we deal with all possible cases.
Table 1. The behavior of the crawler depending on the found links
During page parsing: Check Action
Found link to new
Submit new page to
Record link
Found link to an
already queued or
parsed page

Check if the link:
{father_node, child_node}
is unique
Record link if yes
Ignore link if no
Found other links, not
considered as proper
links (files etc.), or
disallowed by web
- Ignore link

There were several other issues concerning the crawling algorithm that we had to deal with. We provided
flexibility regarding the ability of the web master to block certain types of files or even whole directories of
his site from the crawling process. Consider for example the case where a web site supports more than one
language. Some of those sites implement multilingual services by repeating the whole site with the same
structure, under another directory ( - Such a design
would be mapped by our crawler as repeating the whole link structure one level below the home page, since
the URLs of the pages would appear different for each language and normally the links would not be
identified as already parsed. Cases like the following example would need to be treated specially. We provide
such options for the web masters.
Table 2. Data Retrieved by the Crawler
URL The URL of the link currently parsed.
Father URL The URL of the father node
Title The title of the web page as retrieved from the <title> tag.
Depth The minimum number of steps required to get to this page from the homepage.
Page file The filename of the webpage in the server.
Parent code The html code of the link that lead to this page. Useful for automatic hotlink application.
Directory URL The URL of the pages parent directory in the web server.
Appearance The order of appearance of this specific page, contrast to the appearance of the same page
from other father pages. The greatest Appearance of each page, equals to the number of
different parent pages this page can occur from.
In other equally important issues, we only allow one thread to undertake the crawling, in synchronous
mode, since we noticed rare but existing anomalies in the sequence of the parsed pages when operating in
multi threading mode. Also there was a lot to consider due to the embedment of the crawler to a web
environment in terms of usability and security concerns. In addition, the crawling algorithm underwent
several changes in order to support our need to store several extra data during the crawl. In two different,
intervened stages of the crawling process we collect for each page the data shown in table 2. Finally some
more interventions were made, concerning the parsing of the pages, pages pointing to themselves, URL
normalization and the storing of the collected information in the database.
3.3 Configuring the WebSPHINX Crawler
All the specifications and needs previously described constitute a very different web crawler. Obviously we
needed to start our work on a highly dynamic and configurable machine. The link oriented character
combined with the web environment in which the crawler is embedded, led us to choose a light weight, java
ISBN: 978-972-8939-25-0 2010 IADIS
web crawler named WebSPHINX (Website-Specific Processors for HTML INformation eXtraction) [6]. Java
through its object oriented nature provided us with the perfect base for altering the functionality of the
crawling algorithm as needed. The fact that WebSPHINX is designed for crawling small parts of the web, is
not a problem since our targets are web sites. With careful memory handling, we managed to increase the
crawl capacity up to a few tenths of thousands of pages which is more than enough for our needs.
As aforementioned, our main effort was to make the crawler operate around the notion of the link instead
of the web page - node. In figure 3, we see a small flow chart depicting how our crawler handles the links as
they occur during the parsing of a web page. Notice that each time a link pointing to an already found page
occurs, we increment a counter and attach it to that link. This way we store in the correct order and count all
the appearances of each page. The use of this information will become apparent later on when we present the
Hotlink Visualizer tool. After we check whether the link points to a valid page according to the web
masters criteria, we proceed as described in the first table.

Figure 3. Flowchart that shows the handling of each new link found in a currently parsed page by the crawler.
The crawlers functionality underwent testing both with dummy web sites in the lab and with real, large web
sites on the web. Testing with dummy sites provided answers for targeted algorithmic and functionality
issues and was continuous throughout the specification and development phase. Testing with real web sites
provided feedback concerning size, data storage and memory handling issues and also concluded the
crawlers algorithmic tenability. We also tested with all possible forms of links.
IADIS International Conference WWW/Internet 2010

Figure 4. A dummy web site on the left and the link-structure information that our crawler extracts for it.
(N: Page-node, Fn: Father page-node, D: Depth, #: Counter and order of nodes appearance)
In this section we describe the proposed web, administrative tool Hotlink Visualizer. In short, it is a web
based tool that embeds the crawler described previously and stores in a database all the web sites
connectivity information. It provides a visual representation of the sites structure in a tree-like (expand-
collapse) formation. After storing this map, it gives to the user the ability to add hotlinks to the sites map
with an automated procedure then visualize the outcome and finally make permanent the changes of the link
structure to the site. The option of maintaining different versions of the web sites map is also available. Our
proposed tool guides the user in a step by step process to optimize his site as he or she sees best.
The tool is programmed in Java/JSP which ensures better interoperability with our java crawler,
flexibility and greater functionality in the web context of our study. It is supported by an Oracle 10g Express
Edition database for the data storage, which provides adequate database capacity and scalability when
working with large amount of data. The web server used is an Apache Tomcat 6.0.

Figure 5. The Hotlink Visualizer tool. The web crawlers page for initiating new crawling processes.
The home page of the tool welcomes the user and asks whether he or she would like to start a new crawl,
or work on an already stored instance of the web site of interest. The user beforehand has specified some
information concerning his file system. The paths of the web server and the web applications source files are
required in order to enable the automated hotlink addition. Additional editing paths for the web applications
source code can be specified. In case the user opts for a new crawl, he or she will fill in some information and
N Fn D #
1 A 0 1
2 B A 1 1
3 C A 1 1
4 A B 2 2
5 A C 2 3
6 D C 2 1
7 A D 3 4
8 B D 3 3
ISBN: 978-972-8939-25-0 2010 IADIS
proceed to the next screen that can be seen in figure 5. Here we embed in an applet form the web site crawler
described earlier. The crawler is simplified to the basics however the user still has a number Advanced
options. All the adjustments necessary are set by default, thus all the user has to do is to provide the starting
URL of the web site. Instructions are provided in case the user wants to eliminate parts of his site from the
crawl, or specify unwanted pages or files. The crawler by default looks for pages within the server and only
under the specified homepage. The user during the crawl can see real time statistics for the ongoing crawl.
After the crawling is finished the data is stored to the database and the user can move on to the next step.
The user can get to this next intermediate page of the Available Crawls also from the homepage, if he or
she selects to work on an existing crawl. From this page the web master can see a table containing
information for all the web site instances stored during previous sessions and can also edit basic, descriptive
information about these instances or delete them completely. Whats important is to select from this page the
crawl to work with (from hereon referred to as active crawl), by clicking on the corresponding radio button.
This action activates a specific web site - instance and allows the user to begin interacting with it. A new,
relevant sub menu appears on the right part of the tool titled: Active Crawl: crawl_name, with two options
specific for the selected active crawl. The first option reads Visualize and the second Hotlinks Home.

Figure 6. A small part of a tree-like representation of a dummy web site, crawled by our tool.
By clicking the Visualize link the user progresses to the web sites map visualization page. The
connectivity data stored by the crawler is properly fetched and a tree-like structure is created, depicting the
web site. At first only the home page appears awaiting the expansion, by clicking on the corresponding +
button. Each page, if it has children, can be expanded or collapsed. Some of the web pages-nodes depending
on whether they are hotlinks or whether they belong to the sites direct acyclic graph (consisted of all of the
first appearances of each page) can have appropriate symbols accompanying them. In figure 6 we can see a
small part of a web sites map as generated by our tool. Each page is shown in the sites map is by its <title>
tag, as it is retrieved during the parsing phase. If the title for any reason in not successfully retrieved, the
complete URL is shown instead. The title or URL shown is a physical link to the web page that opens in a
new tab or browser, in order for the user to have immediate views of each page. By clicking on the arrow
next to the title or link of the page, the user is transferred to the Node Details page. The index in figure 6
explains the icons that can be seen in a sites map.
In the detailed page of each link-node the web master can view a full connectivity status concerning the
specific page. Apart from the standard information retrieved during the crawl, such as the URL, depth etc. the
web master can see by a click of a corresponding button a detailed table with all of the other appearances of
the node in the site, another table with all the possible father nodes and also all the children nodes. Finally
from this page the web master can initiate the procedure for a hotlink addition that has as a father page the
current node and as a target a page that he or she will specify next. The addition is completed with the help of
a new side menu that appears when the procedure begins and that guides the user through this quick process.
The Hotlinks Home page is one where the web master can view the hotlinks and their status that have
already been added for the active crawl and perform the automated addition of those already specified but not
yet physically added to the site. The automatic addition uses information about the physical path of the web
sites files in the web server that the user has already specified. The source code of the links that was
retrieved and stored during the crawl is also used in order to provide better formatted links. The links are
appended to the bottom of the father page and though this does not produce the desired result, it usually
simply requires the transfer of the link to the suitable place, thus making the completion of the changes that
the web master aims to be easy, fast and straightforward.
To better illustrate the functionality of the Hotlink Visualizer we present bellow the steps required for a
web master to acquire the exact map of a site, perform the optimizations that he sees best, make them
effective and also maintain the older image of the web site for his own reference and comparison purposes.
1. The user begins by acquiring the initial map of the web site. He does that by choosing to start a
new crawl and then using the embedded crawler as aforementioned.

: Pages 1
appearance (sites DAG)

: Hotlink added by the user

: View pages details
IADIS International Conference WWW/Internet 2010
2. The hotlink assignment procedure is repeated until all the desirable changes are completed.
3. From the Hotlinks Home page the web master automatically inserts those of the suggested
hotlinks approved. The user has the option to specify hotlinks just for visualization and
comparison without adding them. The transformation of the site is completed with this step.
4. Now the user can duplicate the stored instance of his site either automatically or by repeating
step 1 (crawl the altered site and acquire the sites map with the hotlink additions).
5. Finally the web master can activate the first crawl and delete the hotlinks added in step 2.
The above procedure will leave the user with both the new and old versions of the web site. Of course
only the last version exists live on the server. By following the same steps the web master can enrich the
instances-versions of his web site that are stored by Hotlink Visualizer thus providing an archive of site
versions which he or she will be able to edit, compare and apply at any time.
In this paper we attempted an approach on the hotlink assignment problem from a more practical perspective.
We proposed a new administrative tool that aims to help the user exploit this field of research which being so
important, remains relatively unapplied.
The Hotlink Visualizer leaves numerous options open for expansion and future research. First of all, as a
tool oriented towards the World Wide Web, its specifications must constantly evolve accordingly in order to
avoid any limitations in its use. An interesting prospect for future expansion is the crawling in the hidden
parts of the site or the deep Web as discussed in [8]. The potential of embedding invisible crawl functionality
in a web, administrative tool such as the Hotlink Visualizer would broaden massively its perspective and
generality of use. Strictly from a usability point of view, there could be additions, statistic enhancements and
optimizations, depending on the use and acceptance of the tool and of course the feedback. There are already
ideas for a more unambiguous grouping of each crawls hotlinks. What is even more exciting as a prospect
though is to develop a library of hotlink assignment algorithms, suitably specified as to be applicable to the
model of the web site instances that are acquired by our crawler. Research in this direction would require
considerable effort in embedding other hotlink assignment methods in such a library, but will provide the
community with a complete hotlink assignment tool that could be a frame of reference for future web site
optimization research attempts.
1. John Garofalakis, Panagiotis Kappos and Dimitris Mourloukos, 1999. Web Site Optimization Using Page Popularity.
IEEE Internet Computing, July-August 1999.
2. John Garofalakis, Panagiotis Kappos and Christos Makris, 2002. Improving the performance of Web access by
bridging global ranking with local page popularity metrics. Internet Research Westport then Bradford, volume 12,
part 1, pp. 43-54.
3. John Garofalakis, Theodoula Giannakoudi, Evangelos Sakkopoulos, 2007. An Integrated Technique for Web Site
Usage Semantic Analysis: The ORGAN System, Journal of Web Engineering, (JWE), Rinton Press, Vol. 6, No 3, pp.
4. Fuhrmann S., Krumke S.O., Wirth H.C., 2001. Multiple hotlink assignment. Proceedings of the Twenty-Seventh
International Workshop on Graph- Theoretic Concepts in Computer Science.
5. Miguel Vargas Martin, 2002. Enhancing Hyperlink Structure for Improving Web Performance, PhD thesis, Charleton
6. The WebSPHINX crawler main site: Carnegie Mellon University.
7. Junghoo Cho and Hector Garcia-Molina, 2002. Parallel crawlers. In Proceedings of the eleventh international
conference on World Wide Web, Honolulu, Hawaii, USA, pp 124135. ACM Press.
8. He, Bin et al, 2007. Accessing the Deep Web: A Survey. Communications of the ACM (CACM) 50, pp 94101.
9. D. Antoniou et al, 2009. Context-similarity based hotlinks assignment: Model, metrics and algorithm. In press for
journal Data & Knowledge Engineering.
ISBN: 978-972-8939-25-0 2010 IADIS
Rita Oluchi Orji*, Yasemin Yardimci Cetin** and Sevgi Ozkan**
*Computer Science Department, University of Saskatchewan, Canada
**Informatics Institute, Middle East Technical University, Ankara, Turkey
While millions of dollars are spent building digital libraries, research findings indicate that millions of potential users
may still be ignoring them. Researchers have applied different Technology Acceptance Models to determine the
acceptance of Information System (IS). In this research, we recognized the existence of different groups of users of
Electronic Library System (ELS) with different usage behavior and, therefore, developed and validated a Nationality
based Unified Theory of Acceptance and Use of Technology (NUTAUT) model adapted from UTAUT model to account
for each group of users acceptance. Nationality was introduced based on the assumption that the UTAUT independent
variables will impact the acceptance and usage differently when moderated by nationality. The result from 116 (58
international and 58 national) student participants provides support for NUTAUT by showing that the various UTAUT
constructs exert varying degree of effects. It not only confirms the NUTAUT robustness in predicting acceptance of both
National and International Students (91% and 98% respectively) but also determines the importance of each independent
construct to each group in determining acceptance. Performance Expectancy (EE) and Social Influence (SI) are
significant factors for International students while Effort Expectancy (EE) and Facilitating Conditions (FC) have the
same effect for the two groups. Structural Equation Modeling (SEM) was used as the main technique for data analysis.
This study is useful to school managers, bank managers and other Information Science designers that make decision
about IS that are used by people of different nationalities. The NUTAUT model can also be applied in cross-cultural
acceptance and so we provide more insight into the understanding of Technology Acceptance through this study.
Digital Library, Information Retrieval, Technology Acceptance, Structural Equation Modeling
Real world libraries are hard to use. Attempts to reduce the accessibility problems associated with real world
libraries led to the concept of Electronic Library Systems (ELS) (also referred to as, Digital Library). ELS
have become an inevitable part of todays educational systems. ELS aims to acquire, store, organize and
preserve information for easy access. Leedy [16] found that information seekers often need the assistance of
a librarian, especially when the catalogue and guides are difficult to use. In recognition of this, many attempts
have been made towards the establishment and improvement of the structure of a library to achieve both a
high degree of usefulness and easier access to information.
Consequently, many universities have digitized their library systems. However, while many resources
have been devoted to developing these systems, library researchers have observed that digital libraries remain
underutilized [28] and if these systems are not widely utilized, it will be difficult to obtain corresponding
return on investments. Therefore, there is a clear need to identify and compare factors that can influence ELS
acceptance and use by people from different nations so that information system designers, school managers,
library managers and others can formulate strategies to design systems that can be acceptable by all
(international and national). In this regard, Tibenderana and Patrick [23] found that relevance moderated by
awareness plays a major role in acceptance of ELS services. In addition, James et al [13] developed a model
of digital library user acceptance based on the technology acceptance model and concluded that acceptance is
preceded by a users perception of a systems usability and this is directly dependent on interface
IADIS International Conference WWW/Internet 2010
characteristics, organizational contexts, and individual differences. However, no study has recognized the
existence of different group of users of ELS and the factors that affects the acceptance of each group.
In this paper, we argue that national and international users will exhibit different usage behaviors. We
thus, developed a NUTAUT Model to account for the moderating effect of nationality on the UTAUT model.
This was motivated by the observation that different groups of users (national and international students)
exhibit different use behavior towards ELS. Our research model not only predicts the varying degrees of
acceptance for each user groups but also shows the degree of importance of each independent construct
(Performance Expectancy (PE), Effort Expectancy (EE), Social Influence (SI), Facilitating Condition (FC))
in determining acceptance for each group. The study contributes significantly towards understanding of the
acceptance of ELS in academic environments and can also be useful to school managers, bank managers and
other IS designers that make decisions about IS that are used by people of different nationalities.
This paper is organized as follows: In Section 2, we discuss the theoretical background of the study and
present our NUTAUT model in Section 3. Section 4 highlights methods employed in our research and
Section 5 presents the analysis of the result and discussion. Section 6 concludes by presenting our findings
followed by limitations and recommendations for future work.
2.1 Digital Library Systems
There have been significant advances in the technical development of digital libraries in areas such as
information storage, information retrieval, and system integration, resulting in dramatic improvements in
their performance. While many resources have been devoted to developing these systems, library researchers
have observed that ELS remain underutilized [28]. ELS have received a lot of attention from researchers.
Neuman [20] in her naturalistic inquiry detailed some of the difficulties 92 high school freshmen and
sophomore displayed as they interacted with ELS. Her data revealed the basic differences between structures
inherent in database and the conceptual structure that students bring to searching. These differences are so
compelling that they seriously hamper students independent use of these resources (p.74). The students lack
of understanding of the organization of information hampered their access of appropriate information for
their research. The study demonstrated that information search has not become easier with the advent of
2.2 Technology Acceptance Theories
It is a common belief that introducing a new technology automatically results in its acceptance and use.
However, several research findings dispute this claim, showing that there are several other factors that affect
technology acceptance and use [5]. Many IS researcher have published on various theories that could explain
the acceptance of information systems. These theories include the Technology Acceptance Model (TAM) by
Davis et al [7]; the Theory of Reasoned Action (TRA) by Fishbein and Ajzen [9]; the Theory of Planned
Behavior (TPB) by Ajzen, [1] and the UTAUT by Vakentash [26] which is the most recent of all the
technology acceptance theories. The TAM model is the most widely used and has perceived usefulness and
perceived ease of use as its main elements. The model suggests that when users are presented with
technology, perceived usefulness and perceived ease of use influence their decisions about how and
when they will use the technology. The perceived usefulness is defined as the degree to which a person
believes that using a particular system would enhance his or her job performance, while perceived ease of
use is defined as the degree to which a person believes that using a particular system would be free of
effort [7].
Since the development of UTAUT model, it has attracted attention of many scholars in IS research
because of its predictive performance of 70% [26]. This is a major improvement over the widely used TAM
model with predictive capacity of 35% [18, 25]. Various researchers [2, 17] validated the model and others
[6, 12, 19, 23] extended it in different contexts, including multicultural studies [21], and all found its
constructs highly predictive [19]. This, in addition to the fact that the moderating variables offer flexibility to
allow the introduction of new dimensions into the model, was the major motivation for the use of UTAUT
ISBN: 978-972-8939-25-0 2010 IADIS

model in the current investigation. We approached the adoption and discovery of critical factors that affect
adoption of ELS from the perspective of technology acceptance.
3.1 Nationality based UTAUT (NUTAUT)
NUTAUT was adapted from UTAUT by introducing a new modulating variable nationality. The definition
of the NUTAUT constructs is as given in Table 1 and Table 2 and the model is shown in Figure 1.
Nationality was introduced based on the assumption that the UTAUT independent variables PE, EE, SI and
FC will impact on behavior intention (BI) differently and BI with FC will also impact on use behavior (UB)
differently when moderated by nationality.
Table 1. NUTAUT Components
Determinant Description
Performance expectancy (PE) Degree to which an individual believes that using the system will help attain gains in job
Effort expectancy (EE) The degree of ease associated with the use of the system
Social influence (SI) The degree to which an individual perceives that important others believe he or she should
use the new system
Facilitating conditions FC) The degree to which an individual believes that an organizational and technical
infrastructure exists to support use of the system
Behavioral Intention (BI) The measure of the likelihood of an individual to employ the system.
Use Behavior (UB) This measures the acceptance of the system.
Table 2. NUTAUT Moderators
Moderator Description
Gender Gender roles have a strong psychological basis and are enduring.
Age Age has an effect on attitudes.
Experience Deals with how long the user has used the system.
Voluntariness of use If usage is voluntary or mandated
Nationality Whether the user is in the national or international students category

3.2 Hypotheses
The expectations are that there are varying degrees of acceptance by international and national students. Four
hypotheses have been postulated to guide this study:

Gender Experience Voluntariness Nationality
Figure 1. Nationality Based UTAUT (NUTAUT)
IADIS International Conference WWW/Internet 2010
H1: UTAUT moderated by Nationality will demonstrate varying degrees of acceptance by these groups
H2: Social Influence will be an important predictor for international students when moderated by
H3: Facilitating Condition mediated by Nationality will demonstrate more effect on Use Behavior for
international students
H4: Effort Expectancy impacts behavioral intention more than performance expectancy when moderated
by nationality
The data-gathering instrument used for this study was a self-administered online questionnaire. The
questionnaire was based on the pre-existing tool developed by Venkatesh [25] and has been used by
Anderson and Schwager [2], Moran [19], and Tibenderana & Ogao [23]. The research question was divided
into three sections. Section 1 contained 18 close questions which collected the participant demographic
information and their experience with computers and ELS. Section 2 contains a total of 21 questions about
ELS hardware and services provided by Middle East Technical University (METU) library. These questions
collected the students awareness of these facilities and services. The respondents chose either a Yes, No
or Not Sure answer in response to each ELS services and facilities indicated. Section 3 contained 25
questions with a 5-point Likert scale where a 1 represented strongly agree and a 5 represented strongly
disagree. There were a total of 116 participants: 58 International and 58 National graduate students.
Moreover, the number of male and female participants was fairly evenly distributed across the disciplines. A
pilot study was carried out on 10 participants to reassure absence of ambiguity in the questions.
5.1 Validation of the Survey Instrument and Nutaut
The data analysis was done using SPSS and Linear Structural Relations (LISREL) structural equation
modeling tool. SPSS was adopted to conduct Principal Components Analysis (PCA) and to assess the
validity of the scale. The Cronbachs Alpha was calculated to examine the reliability of each factor. The
Alpha values of the questionnaire exceeded 0.8 (Table 3, column 4), demonstrating the good reliability.
Before conducting PCA, Kaiser-Meyer-Olkin (KMO) and Bartlett sphericity test was checked to measure for
the sampling adequacy [14]. The KMO were all >0.700 and the result of Bartlett sphericity test was
significant at <0.001 (Table 3, column 2 and 3). Thus data were suitable to conduct factor analysis [11]. The
factor loadings and the corresponding factor scores (weights) for each variable were generated. Each factor
has larger loading on its corresponding factor (>0.7) than cross-loadings on other factors (<0.4). Thus these
items could effectively reflect factors since they had good validity including convergent validity and
discriminant validity [10].
Structural Equation Model software LISREL was employed to estimate the path coefficients and to
validate and test model hypotheses. We used Confirmatory Factor Analysis (CFA) to test for the model
fitness on the data. The results show that the hypothesized model is recursive, uni-directional (Table 3 & 4).
Also as shown in the Figure 2 and 3, all the standardized loadings of items on their corresponding factor were
larger than 0.7, further proving good convergent validity [4]. The fit indices of the model are listed in Table 4
and Table 5. The tables list the recommended value and actual value of each fit index; the actual value was
better than the recommended value. Thus the model was a good fit to the data. Table 5 summarizes the
results of the t-test analysis which further confirms the validity of the models.

ISBN: 978-972-8939-25-0 2010 IADIS

Table 3. Construct validity and reliability measures

Table 4. Goodness-of-Fit Results of the LISREL models
Recommended >0.90 >0.80 >0.90 >0.90
0.951 0.851 0.946 0.950
0.935 0.901 0.995 0.943

(Note: x2/df the ratio between Chi-square and degrees of freedom, GFI is Goodness of Fit Index, AGFI is Adjusted Goodness of Fit
Index, CFI is Comparative Fit Index, NFI is Normed Fit Index, RMSEA is Root Mean Square Error of Approximation)
5.2 International versus National Students
The two groups result estimations using LISREL are shown in Figures 3 and 4 and the comparison of the
results from the two analyses is summarized in Table 6 and 7. The contributions of the various independent
construct PE, EE, FC and SI for both International and National students are shown in Table 7 column 2 and
3. It shows that FC condition is the most important predictor of acceptance for the two groups. The National
students model shows predictive efficiency of 25% and 73% for the dependent construct of behavioral intent
and use behavior (a total predictive capability of 91% for the dependent variable) and the International
students model shows predictive efficiency of 25% and 66% (a total predictive capability of 98% for the
dependent variable) as shown in Table 7. This means that the two groups accept ELS. This study reveals a
rich set of pattern and results. Due to space limitations, we will only highlight the most important points.

H1: UTAUT moderated by Nationality will demonstrate varying degree of Acceptance by these groups
This hypothesis is supported. There is a significant difference between international and national groups.
National students exhibit a use behavior of 73% and international students exhibit use behavior of 66%. This
shows that national students accept and use the library more than the international students. Though the two
groups demonstrate the same degree of behavioral intent, their results suggest that appropriate facilitating
Group Comparison KMO Bartlett
Recommended >0.5 <0.05 0.7 <0.05 <3 0.08
International 0.758 000 0.879 0.00005 1.31 0.081
National 0.702 000 0.921 0.00007 1.45 0.069
T-test P-Value
National 2.49 <0.001
International 3.20 <0.001
National 4.37 <0.001
International 5.40 <0.001
Table 5. T Test and P Values for participant
Figure 2. National students results estimate Figure 3. International students results estimate
IADIS International Conference WWW/Internet 2010
condition and reduction in the amount of effort expectancy for ELS user enhances the acceptance and use of
technology. This agrees with the finding of Tibenderana & Ogao [23] that library end-users in the university
accept and use electronic library system. The difference in acceptance between the two is mostly attributed to
facilitating condition which is the most important construct for the two groups. A plausible explanation is
that the facilitating condition includes the generality of all the assistance and supports that foster usage rather
than just external controls related to the environment; therefore, use behavior cannot occur if objective
conditions in the environment prevent it [24] or if the facilitating conditions make the behavior difficult. This
is consistent with the empirical studies of Thompson et al [22] and others [25, 26] who found the direct effect
of facilitating condition on usage behavior. International students need to be provided with some additional
assistance which national students might require. This assistance includes training and librarian assistance,
especially at the early stages of their enrolment. This was explained by Neuman [20] statement which states
that there is a basic difference between structures inherent in ELS and conceptual structures that students
bring to searching: these differences are so compelling that they seriously hamper students. Most of the
national participants of this research also did their undergraduate degrees in the university and therefore have
working knowledge of the ELS.
Table 6. Comparison of national and international model contributions
Constructs International(N=58) National(N=58) No. of Questions
No. of Questions
PE 0.50 0.30 4 4
EE 0.52 0.52 6 5
SI 0.12 0.02 3 2
FC 0.78 0.76 6 2
BI 0.25 0.25 3 2
UB 3 2
Table 7. Comparison international student and national students models
Model Behavioral Intention Use Behavior Total
International 25% 66% 91%
National 25% 73% 98%

H2: Social Influence will be an important predictor for international students when moderated by
This hypothesis is also supported; social influence shows a more significant effect on behavioral intent for
international students than national students with a prediction of 12% and 2% respectively. This is expected
since international students are likely to move in groups and therefore are easily influenced to use this system
by people that matter to them in the environment. This can, to an extent, be likened to the effect of
facilitating condition; the groups function as a motivator and can provide necessary assistance especially for
those that are new to the system. Practically, the results of this study suggest that organizations should use
different strategies in motivating the use of a new technology for different situations/groups. For some
information systems whose usage is mandatory, those factors contributing to Social Influence such as the
instructors/supervisor's prodding might work. However, when the usage is voluntary, like the case of ELS,
the managers might want to think of better ways to promote usage. Though social influence is a factor for all
students but show stronger effect for international students, the effect is greatly reduced when moderated by
experience and training, and therefore should not be used as a strong motivator. This is consistent with
previous studies (e.g., Venkatesh & Davis [25]); the effect of SI to BI is significant under mandatory and
inexperienced use situations
H3: Facilitating condition mediated by Nationality will demonstrate more effect on Use behavior for
international students
This is only partially supported with 78% and 76% prediction for international and national students
respectively. This shows that irrespective of nationality, facilitating condition is still crucial. Our initial
assertion that facilitating condition will not be as important for national students owing to the availability of
alternatives including resources from friends and families seems to be wrong. This can possibly be explained
by the fact that the University invests considerable amount of resources to provide both online, offline and
remote access to the ELS, so international students as well as national students still exhibit use behavior on
ISBN: 978-972-8939-25-0 2010 IADIS

the ELS despite availability of alternatives. This is consistent with the empirical studies of Thompson et al
[22] who found the direct effect of facilitating condition on usage behavior
H4: Effort expectancy impacts on behavioral intention more than Performance expectancy when
moderated by nationality
This is partially supported by the model. EE contributed 52% compared to 30% of PE for national
students which support the hypothesis but there is no significant difference between the two for international
student with PE of 50% and EE of 52%. This means that for national students, acceptance is more dependent
on EE than PE, while international students attached similar degrees of importance to both. In other words
national students are unlikely to use the system if it is difficult to use even if it is useful. This can also be
explained by the fact that the availability of alternatives or competition generates a negative effect, affecting
perceived EE. The national students have other sources of getting materials for their research, more easily
than international students, and might not afford to spend a lot of time and energy searching through complex
ELS, while international students value both ease of use and useful systems alike. This could also mean that
in the absence of alternatives, PE becomes as important as EE. This agrees with Andrea and Virili [3]:
Technology acceptance is basically a choice among different alternative technologies/tools to accomplish
user tasks.
The results suggest that Nationality is a significant moderator that affects the acceptance of ELS and that
effect exerted by UTAUT constructs on individual group (international and national students) vary
considerably. Facilitating condition significantly influences the acceptance and use of ELS.
In general, the NUTAUT does predict the successful acceptance of ELS by graduate students. The
participants showed higher inclination to use ELS by showing higher percentage of use behavior. We believe
that this is a result of the role played by readily available access and easy to use library facility on-campus
compared to off-campus, since the majority of the participants stay on-campus and can easily and
conveniently access the ELS as compared to students living off-campus. To increase acceptance of ELS, this
research suggests that the universities working towards increasing acceptance of ELS should make
accessibility (facilitating conditions and effort expectancy) of the ELS to both outside and inside the campus
This study shows that the variables, facilitating condition, effort expectancy, performance expectancy and
social influence (listed in decreasing order of relevance), are the critical components that affect students
acceptance and use of ELS. However, these factors are shown to have varying effect on different groups. For
the two groups (international and national students), facilitating condition and effort expectancy remains the
most critical factors that contribute to acceptance, while performance expectancy and social influences were
fairly insignificant (30% and 2% respectively) for national graduate students.
Finally, though this research focused on the acceptance of ELS in educational institutions, the findings
can be applied to other context like in Government to Citizen (G2C) Systems or websites used for electronic
commerce. These findings can be applicable to any systems where the use and acceptance of the system may
differ by international and national users. The study has contributed to the development of NUTAUT and by
identified the critical factors, in order of importance, that affect acceptance when UTAUT is moderated by
nationality. We also showed that nationality is an important moderator that determines the factors affecting
In conclusion, for a successful implementation of ELS that will be accepted, more emphasis should be
placed on facilitating conditions and effort expectancy but at the same time, performance expectancy and
social influence should not be overlooked.
This research was carried out in a university environment and may not reflect ELS acceptance outside the
university environment, although we plan to validate NUTAUT elsewhere as future work. The number of
participants (116) may indeed be a limitation of this study, and so we want to conduct another study with a
larger number of participants and in another environment to confirm the results shown in this research.

IADIS International Conference WWW/Internet 2010
[1] Ajzen, I. 1991. The Theory of Planned Behavior. Organizational Behavior and Human Decision Processes,50,179-21
[2] Anderson, J. E. and Schwager, P. H. 2006. The Tablet PC: Applying the UTAUT Model, Paper presented at the
American Conference on Information Systems, Acapulco,Mexico
[3] A. V. Pontiggia and Virili, F. 2009. Network effects in technology acceptance: Laboratory experimental evidence.
International Journal of Information Management, 30(1): pp. 68 77.
[4] Bagozzi R P and Yi Y. 1988. "On the evaluation of structural equation models," Journal of the Academy of
Marketing Science, 16(1), pp. 74-94.
[5] Carlsson, C., J. et al 2006. Adoption of Mobile Device/Services Searching for Answers with the UTAUT. Proceeding
of the 39th Hawaii International Conference on System Science. Hawaii. USA. 1 10.
[6] Cody-Allen, E. and Kishore, R. 2006. An Extension of the UTAUT Model with E- Eqality, Trust & Satisfaction
Constructs. CPR: 183-189.
[7] Davis, F. D. et al 1989. User Acceptance of Computer Technology: A Comparison of Two Theoretical Models.
Management Science. 35(8): 982-1002.
[9] Fishbein, M. and I. Ajzen, Belief, 1975. Attitude Intention and Behavior: An Introduction to Theory and Research.
Reading, MA: Addison-Wesley Publishing Company.
[10]Gefen D et al 2000. "Structural Equation Modeling and Regression:Guidelines for Research Practice,"
Communications of the Association for Information Systems, vol 4, pp. 1-70.
[11]Guo Z. 1999. Social Statistics Analytical Methodsthe Application of SPSS. Beijing: China Renmin University Press
[12]Heerik, M. et al 2006. Human-Robot User Studies in Eldercare: Lessons Learned. Institute for Information
Engineering, University of Amsterdam, Oudweg, The Netherlands.
[13] J. Y. L Thong, et al 2004. What leads to user acceptance of digital libraries? Communications of the ACM, 47(11),
[14]Kaiser, H. 1970. A Second Generation Little Jiffy. Psycometrika, 35, 401-415.
[15]Lee, Y. et al 2003. The Technology Acceptance Model: Past, Present, and Future. Communications of the
Association for Information Systems. 12(50): pp.752-780
[16]Leedy, P. 1993. Practical Research, Planning and Design: Columbus, Ohio, Merrill, 17-42.
[17]Louho, R., M. et al 2006. Factors Affecting the Use of Hybrid Media Applications. Graphic Arts in Finland. 35: 3.
[18]Meister, D. B. and Compeau, D. R. 2002. Infusion of Innovation Adoption: An Individual Perspective: Proceedings
of the ASAC. Winnipeg, Manitoba.
[19]Moran, M. J. 2006. College Students Acceptance of Tablet PCs and Application of the Unified Theory of
Acceptance Technology (UTAUT) Model. Ph. D. Thesis. Capella University. Mennesota, USA.
[20]Neuman, D. 2004. Learning and the Digital Library. Youth information seeking behavior. Theory, model and issues
(pp. 65-95).
[21]Oshlyansky, L. et al 2007. Validating the Unified Theory of Acceptance and Use of Technology (UTAUT) tool
cross-culturally. Proc. of HCI. 2 BCS, 83-86.
Pontiggia A. V. and Virili, F. 2009. Network effects in technology acceptance: Laboratory experimental evidence.
International Journal of Information Management, 30(1): pp. 68 77.
[22]Thompson R. L. et al 1991. Personal Computing: Toward a Conceptual Model of Utilization. MIS Q. 15(1):12543.
[23]Tibenderana, P. K. and Patrick J. O. 2008. Acceptance and Use of Electronic Library Services in Uganda
Universities. ACM JCDL.
[24]Triandis, H. 1980. "Values, attitudes, and interpersonal behavior", Nebraska Symposium on Motivation, pp.195-259.
[25]Venkatesh, V. and Davis, F.D., 2000. A theoretical extension of the technology acceptance model: four
longitudinalfield studies. Management Science, 46 (2), 186204.
[26]Venkatesh, V. et al. 2003. User Acceptance of Information Technology: Toward a Unified View. MIS Quarterly,
27(3): 425-478
[27]Wiederhold, G. 1995. Digital libraries, Value and Productivity. Communicaton of ACM 38(4) 8596.
[28]Wood, F. et al, 1995. Information Skills for Student Centered learning. The Association for Information
Management. Milton Keynes, London, UK, 134148.
[29] Wixom, B. H. & Todd, P. A. 2005. A Theoretical Intergration of User Satisfaction and Technology Acceptance.
Information Systems Research, 16(1): 85-102.
ISBN: 978-972-8939-25-0 2010 IADIS
Paulo Caetano da Silva*, Marcelo Mendona dos Santos** and Valria Cesrio Times**
*Federal University of Pernambuco - Center for Informatics, PO BOX 7851, Brazil
**rea 1 Science and Technology Faculty, Salvador, Brazil
XLink are XML elements used for describing relationships and semantic information contained in valid XML documents.
Thus, the structure of XLink can be navigated and their components (arcs, locators, resources and title) can be retrieved
through a path language, like XPath. However, it was not specified to retrieve information from specific elements of
Xlink, which can contain semantic information. This paper proposes XLPath, a path language turned to navigation on
XLink, which allows performing queries on the elements that compose the links between XML documents. Aiming to
demonstrate the use of XLPath, a processor prototype for this language was developed, which was evaluated in a case
study based on XBRL documents, since XBRL is a technology that makes extensive use of XLink.
XLPath, XLink, Navigational Language, XML.
Current software applications usually have to deal with multiple data sources and data formats. In order to
minimize this problem, XML has been adopted as a way to integrate data in a standard format. As a result of
the increasing need of data integration and data exchange, XML documents are turning into very huge and
rather interlinked files. Often, these documents have complex link networks pointing to all kinds of
resources. These resources should be used only combined with the XML document, since their semantics are
defined by the link networks. The XML Linking Language (XLink) [XLink, 2001] is used to describe
relationships among resources included in XML documents and referenced by links. By the use of XLink, it
is possible to create association rules among the link resources, as well as to define the semantics needed for
traversing a given link according to the specific meaning of this relationship. Processing documents with link
networks has become a challenging task, because query languages usually do not support link traversing
One of the mechanisms used on data structures based on XML are the extended links provided by XLink.
The extended links allow the establishment of more complex link structures, capable of relating an arbitrary
number of resources [Lowe, 2001]. An extended link basically consists of an XML element that contains
other elements, where a namespace [Namespace, 2009] is used for declaring attributes specified by XLink,
assigning to these sub-elements functionalities. Due to the functionalities offered by these mechanisms, in
some situations the extended links have been used to represent the data semantic, modeling relationships
between elements structured on XML or on XML Schema [Schema, 2004]. An example of how the data
semantic is represented on XLink is XBRL (eXtensible Business Reporting Language) [XBRL, 2003], an
international standard for represent and publish financial reports that uses extended links for modeling
financial concepts
However, despite of the functionalities provided by XLink, the maintenance and manipulation of
structures that are based on extended links becomes a costly and inefficient task when a solution to automate
the navigation through these interconnected data structures is not available, making difficult the execution of
data retrievals from the extended links. In this scenario, a process that demands the identification of remote
information, from an extended link, implies in a need of users performing a detailed analysis of the XML
document structure, navigating visually through the references established by the links until they reach the
desired information. Obviously, as the volume of data stored in the database increases, this navigation task
IADIS International Conference WWW/Internet 2010
may become prohibitive. A possible solution would be the use of XQuery [XQuery, 2007] or XPath [XPath,
1999] for recovering information structured on XML. The first one consists of a query language for XML
data, just like SQL is for relational data, while the second one allows exploring the nodes of the tree based
structure of XML documents. According to [Harold et. al., 2004], XPath is a language to capture nodes and
nodes groups of this tree. However, as we will discuss in Section 2, these languages appear to be insufficient
when interconnected XML data are queried. It demands, therefore, a language whose properties are
specialized on the treatment of this link network. In Section 6, we will investigate the solutions created with
this purpose. Nevertheless, we will show that the deficiencies and restrictions presented by them justify the
creation of the language proposed in this paper.
In this paper, we propose XLPath (XML Linking Path Language), a language that is based on query
expressions, is similar to XPath, and has been designed for navigating in an extended link network of XML
documents. This language allows the elements that compose the networks to be accessed according to
refining criteria defined by users. This paper is organized as follows. In Section 2, the concepts about
extended links and the limitations of XQuery and XPath to manipulate these links are presented. The
specification of XLPath language is presented in Section 3 and its processor is detailed in Section 4. A case
study based on a XBRL document is discussed in Section 5. In Section 6, corelated studies are evaluated
regarding their limitations and advantages. At last, this paper conclusion is in Section 7.
In this section, we will discuss concepts inherent to extended links, showing an example, Figure 1, which will
be used through the whole paper to explain XLPath. It will also be discussed the restrictions that happen in
current W3C [W3C, 2010] standard languages to the navigation and query in XML documents.
XLink provides two types of links. The first one, the simple type, may be considered as a simplification
of the extended link. A simple link associates exactly two resources, one local and one remote, with an arc
going from the former to the latter. The second type corresponds to the extended links, which allow the
connections creation: (i) outbound in which the source resource is local and the destiny resource is remote;
(ii) inbound when the source resource is remote and the destiny resource is local; and, (iii) third-party
when both, source and destiny, are remote. In this case, separated documents are created, called linkbases, in
which the links are grouped. Using extended links it is possible to associate an arbitrary number of resources.
In order to provide such flexibility, the structure of extended links contains elements that: (1) point to remote
resources through an URI xlink:type=locator; (2) consist of local resources - xlink:type=resource, used
for encapsulating information inside the link; (3) define arc traversing rules - xlink:type=arc. An arc
connects two resources and provides information about the link traversing, such as the navigation direction
and the application behavior regarding the traversing; and (4) provides descriptive information regarding the
link - xlink:type=title. In this paper, the terms locator, resource, arc and title refer to the homonym type
Figure 1 illustrates a code excerpt for an extended link of the type third-party. This code was extracted
from the XBRL taxonomy of project COREP [COREP, 2005], an initiative of CEBS (Committe of European
Banking Supervisors) to provide a framework of financial reports for some institutions from European Union.
In this code, locators were declared to refer the elements d-ba_BankingActivitiesDomain, d-
ba_TotalBankingActivities and d-ba_TotalBankingActivitiesSubjectBIA, found in the scheme d-ba-2006-07-
01.xsd. In addition, arcs are used to establish relationships of the type domain-member between these
elements. In virtue of the increasing usage of XML as a mean for storing data, query languages carry out a
base role, allowing access to data and metadata that compose the semi-structured data files. However many
languages with this purpose exist, as XQL [Robie et al., 1999] and XML-QL [Deutsch et al., 1998], the
languages constructed by W3C, XQuery and XPath, attend the main needs of access to databases and XML
documents. The main use of XQuery is the performance of queries to data stored in native XML data bases.
The XPath use query expressions called location path, which determine a navigation path between the
document parts, called nodes, exploring the relationships characteristic of the tree formed by these nodes (e.
g. parent, child, sibling). In extended links case, once they are XML base data, it is possible to explore them
under the nodes tree perspective through XPath. Figure 2 shows a simplified nodes tree (without the
attributes) that characterizes the extended link of Figure 1.
ISBN: 978-972-8939-25-0 2010 IADIS

Figure 1. Example of an extended link.

Figure 2. Extended link nodes tree.

As an example of XPath expression, the following location path may be used to find, from the XML
document root, the element locator whose label is BankingActivitiesDomain:
/descendantorself::link:definitionLink/child::link:loc[@xlink:label=BankingActivitiesDomain]. The query resulting from
the computation of this expression is performed in stages called location steps (isolated by /), whose
general form is: axis::nodeSpecification[Predicate].
The axis indicates in each direction from one reference node called context node the query must be
performed, placed that in majority of cases this direction regards to one of the relationships of the XML tree.
The node specification determines which nodes in this direction must be selected. Besides that, it is possible
to filter the result using a predicate (optional), which consists in a boolean test applied to the selected node:
when the result is true, the node is maintained; otherwise, it is discarded. The expression organization in
successive location steps allows any node in the tree to be accessed, making XPath an ideal tool for
navigating in this type of structure. XPath has been seen as a de facto standard in the XML query research
area. However, it does not provide a means of navigating through XLink links. As a result, both the
semantics and the processing issues concerning link data are compromised.
When the interest is exploring the information according to the extended links perspective, this solution
appears to be insufficient. For example, to formulate a query whose purpose is to select all the destiny
elements of arcs, whose source is the element d-ba_TotalBankingActivities, two important issues must be
considered: (i) differently from a traditional XML tree, the data structure resulting from references
established by the links characterize a network; and (ii) there is no restriction regarding the location of a
referred resource, which may be in any part of the XML document or, yet, in a distinct document. Thus,
exploring the information related by the extended link using XPath, when it is not possible (in the case of
refreences between distinct XML documents), demands a formulation of complex queries. With XQuery the
situation is not different, since mechanisms as FLWOR clauses are not proper for navigation in links. It is
evident, therefore, the need for a query language specialized in the extended links mechanism. However, it is
important to ponder that the development of a language based on a completely new syntax would cause a
bigger difficult assimilation from the users who are already familiar with XPath syntax. To minimize this
impact, it is desirable the existence of a proximity between these syntaxes. The language proposed in this
paper, XLPath, is based on location path expressions of XPath, in which new axis, node tests and predicates
were developed aiming the navigation through the networks formed by extended links.
child parent
IADIS International Conference WWW/Internet 2010
In this section, the language proposed will be presented. In Section 3.1, the requirements identified as a guide
to XLPath specification are given. In Section 3.2, we will discuss two approaches to perform queries used in
XLPath. In Section 3.3 the language syntax and semantic are detailed.
3.1 XLPath requirements
From the need verified regarding the navigation over links, as well as XLink conceptual approaches, it is
possible to compile a series of requirements have guided the XLPath construction: i) Conceptual distinction
between arcs and connections, and the elements that define them From XLPath point of view, an arc or
a connection must be taken as an act of an element refer another, through a label present in the attributes
from and to of XLink, or through an URI in the href attribute. Thus, it is correct to affirm that, when
regarding queries based on this language, an arc may start from an element locator to an element arc, or vice-
versa.; (ii) Navigation in different levels of abstraction The main purpose of a query language over links
is allowing the navigation through the network formed by the existing references between distinct elements.
However, references through URI (href) constitute the only possible path between two elements that refer
themselves in this way. Thus, XLPath must provide the option of abstracting these connections, decreasing
the quantity of steps necessary to reach the query final purpose; (iii) Approach for simple links as their
extended equivalents Aiming to avoid a very long syntax, XLPath must be capable of navigating over
simple links based on the same syntax used for extended links. To do so, this language must start from an
approach that assimilates the simple links as their extended equivalents; (iv) Identification of implicit arcs
In situations in which links are declared without the occurrence of arcs, or that, in the occurrence of them, the
attributes from and/or to have been omitted, XLPath must be capable of identifying the occurrence of implicit
arcs, maintaining, thus, the conformity with the XLink specification; (v) Similarity with the XPath syntax
The language proposed here for navigation over links must have the maximum possible number of
similarities with the XPath syntax, in order to facilitate the assimilation for the users. (vi) Conditions for
queries refinement Placed that may exist an arbitrary number of links referring a given element, XLPath
must enable refinement conditions to be applied to the query, in order to allow the distinction among the
different elements that constitute these links, starting from their names or values of attributes; and (vii)
Absolute and relative path, and query performed in steps J ust like it happens in XPath, XLPath follows
the concept of query in steps, where each step selects a list of nodes that is passed as an input parameter for
the next step until the execution of the last step and obtainment of the final result. The initial step may start
from a document root (absolute path), or from a specific node in its interior (relative path).
3.2 Graphical view
One way of analyzing how the structure formed by extended links may be explored by a query language is
representing it graphically. We suggest two different approaches to represent the links, Low Level Approach
and High Level Approach. These approaches are exemplified through a link illustrated in Figure 1, adopting
the symbolism presented in Table 1.
Low Level Approach: The low level approach offers a detailed view of the link, considering all the
elements involved in the existing references, which includes locators, arcs and resources. The advantage of
this approach for a query language is to allow the access to all these elements. Figure 3 shows the link of
Figure 1 represented in low level. To differentiate elements of the same kind and with the same name,
numbers corresponding to the order in which each element appears in Figure 1 are used as labels.
High Level Approach: In an extended link, a type locator element has the function of being the local
representation of certain remote resource. Due to it, for each remote resource there will be only one locator
referring it. The high level approach offers a simplified vision of the link, where the locators are omitted and
the arcs make direct reference to the correspondent remote resource. For a query language, it would result in
the formulation of simpler queries to do a link crossing (when we navigate from one resource to another
passing by the link). Figure 4 shows the representation in high level for the link shown in Figure 1.
ISBN: 978-972-8939-25-0 2010 IADIS
Table 1. Simbology for extended links.
Simbol Representation






Figure 3. Low level representation of an extended link. Figure 4. High level representation of an extended link.
3.3 Syntax and semantic
This section presents the XLPath sintactic mechanisms, accompanied of examples that demonstrate their
interpretation. The basic structure of an XLPath query expression is similar to the location path of XPath,
including the axis elements, element specification analogue to the node specification of XPath and
predicate. Besides those, to enjoy the advantages of the two navigation approaches suggested in Sections 3.2,
the XLPath query expressions make use of a mechanism that indicates the level of abstraction. Thus,
expressions preceded by LL navigate through the extended links structures according to the low level
approach, while expressions preceded by HL navigate in high level.
Axis Specification: The basic blocks of an expression location path of XPath are the location steps, which
determine the paths that conduct to the desired information in an XML document. These paths are directed by
the axis, which, in XPath, correspond mainly to the traditional relationships between nodes of XML
documents trees. In XLPath, the same principle is used, however with the difference of the paths, in this case,
are determined by the references among the elements that compose the links networks. Graphically, these
references correspond to the lines and arrows that connect the elements as given in Figure 3 and in Figure 4.
It is possible to realize that elements of the type arc are always associated to two link references: one
through the attribute xlink:from and another through the attribute xlink:to, each one of them represented by
an arrow, which identifies the directional character of this type of association. An important detail is that,
from XLPath perspective, each one of these arrows is considered an arc. It means that there is a distinction
between arcs and elements of the arc type. This distinction is made to allow that XLPath access elements of
the arc type like nodes of the links networks, navigating until them through arcs, i.e. references originating
from elements of the arc type. Thus, XLPath axes are used to specify the type of existing association between
the context element equivalent to the context node of XPath and the elements connected to it. These axes
linked axis directs the query to all the elements that refer or are referred by context element,
including references by label and by URI;
arc-source axis directs the query to all the elements that are target nodes of arcs whose context
element is the source;
4 6
IADIS International Conference WWW/Internet 2010
arc-destination axis directs the query to all the elements that are source nodes of arcs whose context
element is the target.
The first axis, linked, is the most generic. Graphically, the result of its application is equivalent to
propagate the query through all the lines and arrows connected to the context element. The other axes, arc-
source and arc-destination propagate the query through arrows whose context element is source or target,
respectively. Table 2 exemplifies the axes application, in different levels of abstraction, to the elements of
Figure 1. To differentiate elements with repeated names, the same numeration used as labels in Figures 3 and
4 were superscribed on them.
Table 2. Examples of abstraction levels and axis. Table 3. Examples of element specification.
Contextelement LevelandAxis IdentifiedElements Contextelement ElementSpecification SelectedElements
locator() link:loc
LL::arcsource link:definitionArc
6 link:definitionArc


HL::arcsource link:definitionArc
6 LL::arc
Element Specification: The element specification determines which elements in the axis direction will be
selected. The selection criteria may be the type and the name of each element, determined, respectively, by
the parameter of the function elementType(elementName). XLPath provides the following element
locator(elementName) selects elements of locator type with the name elementName;
arc(elementName) selects elements of arc type with the name elementName;
resource(elementName) selects elements of resource type with the name elementName;
remote(elementName) selects remote resources with the name elementName;
element(elementName) selects elements of any type with the name elementName.
When the parameter elementName is not provided, the filtering is made based only on the type of
element. Thus, the function element() may be used to put all the elements back in the direction of an axis,
regardless the name and the type. Table 3 shows examples of element specification applied to elements
showed on Figure 1.
Predicate: In the examples of Table 3, in some situations, even specifying the element name and type, the
result is an element list. To filter this result, one may use predicates, which consist in conditions tested on
each element on the list. J ust like in XPath, the usage of predicates is not mandatory. Table 4 exemplifies the
predicates to filter the list of elements extracted from Figure 1. XLPath provides the following conditions:
attribute(attributeName)=attributeValue tests if the element has an attribute attributeName with
the value attributeValue;
text()=textValue tests if the text content of the element equals textValue;
link(linkName) tests if the element belongs to the link linkName.
Table 4. Examples of predicates
Contextelement Predicate FilteredElements


link(link:definitionLink) link:loc

The XLPath grammar is illustrated in Figure 5, described through representation based on an EBNF [37].
ISBN: 978-972-8939-25-0 2010 IADIS

Figure 5. XLPath EBNF.
This section presents the definitions of an XLPath language processor, whose architecture is based on MVC
(Model-View-Controller) pattern. This architecture and the processing algorithm are described follow.
4.1 Architecture
Following MVC pattern, the Model layer of XLPath processor is formed by the components: (i) Expression
maintain XLPath query expressions; (ii) Step sub-divide XLPath expressions in steps to be sequentially
processed; and (iii) Element Store store the elements resulting from each step processing.
The layer View has the components: (i) Console provides the user with an interface for the entrance of
query expressions and documents, as well as with the presentation of the obtained results; and (ii) Formatter
format the results for visualization by the user.
The layer Controller is composed by the modules: (i) Executor generate a representation of the input
documents in the memory, based on DOM (Document Object Model) technology [Hgaret et al., 2005], and
conducts it to the processing together with XLPath expression; and (ii) Processor processes the XLPath
expression steps sequence and stores the result in an Element Store.
Figure 6 brings a representation of XLPath processor containing the components described here.
4.2 Algorithm
XLPath processing algorithm is based on the functions xlpathExecutor and xlpathProcessor. At first,
xlpathExecutor receives as a parameter an XLPath expression and a representation in DOM of an XML
document. From them, it identifies the initial context element and invokes xlpathProcessor, passing as a
parameter the context element identified and the query expression. Next, xlpathProcessor interprets each step
of XLPath expression, going through the links network to identify the interest referentiations that involve
each context element. Figure 7 illustrates this algorithm.
The query used as an example in Section 2.2, besides other examples, is solved bellow through XLPath:
Select all arcs destiny elements whose element d-ba_TotalBankingActivities is the origin.
Select the locator belonging to the link:definitionLink that connects itself to the element d-
XLPathExpr ::= AbsLocationPath|RelLocationPath
AbsLocationPath ::= /RelLocationPath
RelLocationPath ::= Step|RelLocationPath/Step
Step ::= LevelSpecAxisSpecElementSpecPredicate*
LevelSpec ::= (HL|LL)::
AxisSpec ::= (arcdestination|arcsource|linked|linkpartchild|linkpartdescendant)::
ElementSpec ::= ElementName(QName?)
ElementName ::= arc|element|locator|remote|resource
Predicate ::= [OperatorSpec]
OperatorSpec ::= AttributeSpec|LinkSpec|TextSpec
AttributeSpec ::= attribute(QName)=Value
LinkSpec ::= link(QName)
TextSpec ::= text()=Value
IADIS International Conference WWW/Internet 2010
Select the remote resource that attend the following conditions: (i) be connected to the locator whose
label is TotalBankingActivities; and (ii) be the origin of the arc whose element db-
a_TotalBankingActivitiesSubjectBIA is destiny.
Execute the link crossing from the element d-ba_BankingActivitiesDomain to the element d-

Figure 6. XLPath processor architecture. Figure 7. XLPath algorithm.
These examples show how XLPath perform queries in XML documents interconnected by XLink. To
demonstrate its usage in a real application, a prototype for XLPath processor was developed in J ava
language, based on the architecture and algorithm defined in Sections 4.1 and 4.2. Each component of this
architecture was modeled in a J ava class, and these classes were grouped in packages named Model, View
and Controller, according to MVC approach. Figure 8 shows the XLPath processor console. This prototype is
composed by two input data parameters one for the query expression and another for the document to be
queried , besides a field for data output, where the query results are exhibited. Figure 8 illustrates a query
performed in the document d-ba-2006-07-01.xsd, Figure 1. This document contains elements referred by the
link link:definitionLink, which is in a linkbase. During the query execution, the XLPath expression identifies
these references and, automatically, identifies the corresponding linkbase. This type of query cannot be
performed by the languages XPath, XQuery, or by the proposals discussed in Section 6. Thus, of the main
contribution of XLPath is the ability to perform queries in XML documents interconnected by extended links.
The output data of an XLPath query is organized by the XLPathFormatter class, which shows the steps
sequency of the XLPath expression submitted and the respective resulting elements of its processing by the
XLPathProcessor class. Figure 9 illustrates XLPath query based on the document illustrated in Figure 1and
the complete result extracted from the console. In this example, XLPath expression submitted was composed
by five steps, to combine low and high level approaches, and, in some cases, making use of predicate
operators to refine the query. Thus, all the mechanisms provided by XLPath could be tested: the specification
of abstraction level, of axis and of element, and the predicate operators.
XLPath comes from XPath+language [Silva et al., 2008 and 2009], which main characteristic is allowing
queries fulfillment over the arcs that compose the links. This language is capable of locating arcs, whether
they are in an instance, in an XML Schema data scheme, or in a linkbases. The main mechanisms provided
by XPath+are the link-source and link-destination axis. The first selects a list of nodes that has context node
as a source resource, while the second selects a list of nodes that has the context node as a destiny resource.
This semantic is reproduced in XLPath through the arc-source and arc-destination axis, however combined
with other mechanisms that allow the filtering of the elements obtained as a result of its application.
Console Expression


Controller Model


Formatter Element

ISBN: 978-972-8939-25-0 2010 IADIS

Figure 8. XLPath processor console.

Figure 9. XLPath processor exit.
May and Malheiro [2001 and 2002] propose an extension for the namespace xlink, dbxlink, which
enables links behavior specification in the moment of the query. Through the inclusion of namespace
dbxlink, it is possible to choose how the linkbase elements are added to the original XML document.
However, to do so, the XML document must be changed, including at least, the standard setting
dbxlink:transparent. This is not very practical in real applications and it is not applicable in documents whose
relationship semantic is important. Because of this semantic in the linkbases, what is expected is the recovery
of the meanings without changing the documents. May et al. [2008] propose an implementation to perform
queries in XML based on dbxlink. The links are not seen as explicit connections, in which the users must be
aware of the links and explicitly cross them in their queries. This proposal extends the XML native data base
system eXist [eXist, 2009]. Different from XLPath, which proposes itself to be a language for navigation and
query to XML documents interconnected by XLink, this solution is included in the category of XML data
base system. Besides, the XLPath proposal is based on following the links explicitly when consulting and
declaring them explicitly in its operators.
IADIS International Conference WWW/Internet 2010
Bry and Eckert [2005] also propose and extension of XLink for linkbases processing in the Web. They
introduce the concept of interface for links structure and a linkbases management mechanism. This proposal
is different from XLPath in the purposes, because, while XLPath is proposed as a navigation language for
links based on elements and attributes defined in XLink, Bry and Eckert proposal suggests a way of
managing linkbases in the Web. The proposal described by Lizorkim [Lizorkim et al., 2003 and 2005, and S-
exp, 2009] suggests an extension of XPath. Despite of this solution be similar to XLPath, because it is an
extension of XPath, its implementation is performed based on expressions S, a native type in the functional
language Scheme. This requires the conversion of XML files to the SXML format, a representation of XML
Information Set in S-expression form. The SXPath proposal refers to the implementation of the XPath
language to manipulate links in the functional language Scheme. Thus, in the SXPath approach, there is
initially the conversion of the XML document to SXML. Like occurs in XLPath, it is possible to highlight the
coherence that exists between the data structuring form (DOM and S-expressions) and their manipulation
(J ava and Scheme). However, to reach this coherence, SXPath needs to convert the XML format in SXML,
which, in spite of being similar, results in an additional effort. Ahmedi [2001 and 2005] proposal was based
on the protocol LDAP (Lightweight Directory Access Protocol) to make a query, through XPath expressions,
XML documents interconnected with links of the type XLink. As LDAP has a data model different from
XML, XML documents with the links need to be mapped for this model. A similar solution, proposed by
Ahmedi and Arifaj [2008], there is no need for mapping for a data model different from XML. However,
XPath queries, in this proposals, are based on the xlink:href attribute, not exploring all the power of semantic
expression of XLink, i.e., the queries processing do not consider the semantics attributes of XLink (title, role
e arcrole), the behavior attributes (show e actuate), or the values for the xlink:type attribute (resouce, locator,
arc and title).
XLink Processor (XLiP) is a Fujitsu implementation [XLip, 2005] to extract XLink and XPointer
information in XML documents. Based on Java and DOM, enables the use of all types of links expressed by
XLink. However, diferently from XLPath, it is owner. Other commercial solution to deal with XLink, based
on J ava and DOM, it is Batavia XBRL J ava Library (BXJ L) [Batavia, 2009], it is however specific to
manipulate only XBRL documents. J ava XBRL API Implementation [XBRLAPI, 2008] is an open code
project that provides an XLink processor for XBRL documents. This processor is a solution based on SAX,
manipulating XLink events. Another approach based on SAX is XLink Filter [Laurent, 1998], which creates
a links collection that may attend to applications requires to identify which elements contain links, as well as
their purposes and behaviors. SAX usage has the advantage of being more efficient regarding memory use,
but is not a solution standardized by W3C. XLinkit [Nentwich et al., 2000] is a tool to generate links
according to rules and verify the documents consistence. As an entrance, the tool receives a group of XML
documents and a group of potential rules that connect these documents contents. The rules represent
consistence restrictions among the resources. XLinkit conduct back a linkbase with a group of links XLink,
which allow the navigation among the resources elements.
In XSPath [Cavalieri et al., 2008], the notion of queries performed through steps, present on XPath
expressions, is extended in order to allow the navigation through the connections contained in the scheme
that associates the elements, specified in XML Schema documents, to their data types, being also possible to
navigate inside data complex types [W3, 2010]. However, there is not always the interest in explore a data
complex structure. Therefore, XSPath offers two different navigation approaches: low level expressions,
which analyze complex types intern structures, and high level expressions, which steps do not depend on how
the element are combined to form complex types.
According to the study of these proposals, to attend the navigation in XML documents interconnected by
links, some solutions require a transformation in the XML document, as the proposals based on SXML, the
LoPiX, and the one which defines three navigation modes in linkbases (transient, temporary and permanent),
others are limited to specific solutions, such as the ones for XBRL and the XLinkit. J ava XBRL API
Implementation and XLink Filter are funded by the use of SAX, what restricts its application, since SAX is
not a W3C pattern. Besides, none of these solutions are specified as language for navigation in links of the
XLink type. XSPath is a proposal of language for navigating in XML Schema documents, which served as a
base of the high and low level approaches for XLPath queries proposed in this paper. Table 5 presents some
results derived from the comparison among the proposals discussed in this section and XLPath. This analysis
is made considering the capacity for links manipulation. It is settled that XLPath manipulates any type of link
based on XLink and that it does not require any change on the original XML documents. Besides, XLPath is
an open code solution and based on patterns W3C (DOM, XPath, XML Schema and XLink.) and it is a
ISBN: 978-972-8939-25-0 2010 IADIS
navigation language for XLink. These characteristics may be considered as advantages of XLPath compared
to other approaches.
Table 5. Comparison among XLPath and other proposals.
Proposals ChangestheXML
Opencodesolution Languagefornavigation
dblink Yes No Yes Yes No
SXML/SXPath Yes No Yes Yes Yes
XlinkFilter No No Yes Notinformed No
XLiP No Yes Yes No No
BataviaXBRL No Yes No No No
JavaXBRLAPI No No No No No
Xlinkit Yes Yes Donotapply Notinformed No
ManagementofLinkbases Yes No Donotapply Yes No
XLPath No Yes Yes Yes Yes
The language for navigating over links proposed in this paper allows the performance of queries with
expressions that are similar to XPath, then the paths formed among the elements that compose the links are
gone through, and these elements accessed. Besides that, the language provides means to establish criteria
according with the navigation will happen, enabling this way that users and/or applications reach specific
elements that compose the extended links networks. XLPath uses mechanisms based on XPath, such as axis,
element specification and predicate. XLPath uses axes that guide the navigation according to the context
element condition, as arcs source or destiny. This semantic is based on the axes originating from XPath+
language. In XLPath, the mentioned mechanisms are used to compose navigation expressions, whose
arrangement is similar to the location path expressions, from XPath. Thus, its syntax assimilation is
facilitated, above all for those users who are already familiar with XPath. In order to continue this project, it
is intended to develop a new version for XLPath processor, which would allow the navigation through links
present in distributed data sources. The addition of an optimizer module in the XLPath processor is seen as
another indication of future work.
Ahmedi, L. and Arifaj, M., 2008. Processing XPath/XQuery to be aware of XLink hyperlinks. 2nd European Computing
Conference (ECC08), Malta, pp. 217-221.
Ahmedi, L. and Arifaj, M., 2008. Querying XML documents with XPath/XQuery in presence of XLink hyperlinks.
WSEAS Transactions on Computers, Vol. 7, No. 10, pp. 1752-1751.
Ahmedi, L., 2001. Global access to interlinked XML data using LDAP and ontologies. PhD dissertation fromAlbert-
Ludwigs, Freiburg.
Ahmedi, L., 2005. Making XPath reach for the web-wide links. ACM Symposium on Applied Computing. Santa Fe, US,
pp. 1714-1721.
Batavia Business Reporting. Batavia XBRL Java Library (BXJ L). Retrieved J anuary, 2009, from SourceForge:
Bry, F. and Eckert, M., 2005. Processing link structures and linkbases in the webs open world linking. Proceedings of
sixteenth ACM Conference on Hypertext and Hypermedia, Salzburg, AT.
CavalieriI, F., Guerrini, G. and Mesiti, M., 2008. Navigational path expressions on XML schema. 19th International
Conference on Database and Expert Systems Applications DEXA, Turin, IT.
COREP XBRL Project, 2005. Retrieved J uly 10, 2010, from Eurofiling:
Deutsch, A., Fernandez, M., Florescu, D., Levy, A. and Suciu, D., 1998. XML-QL: a query language for XML.
Retrieved J uly, 2010, from:
Exist. eXist-db Open Source Native XML Database. Retrieved September, 2009, from SourceForge:
Harold, E.R. and Means, W.S. 2004. XML in a nutshell. O'Reilly, Sebastopol, US.
IADIS International Conference WWW/Internet 2010
Hgaret, P.L., Withmer, R. and Wood, L., 2005. Document object model (DOM), 2005. Retrieved August, 2010, from:
Laurent, S.S., 1998. XLinkFilter: An open source J ava XLink SAX parser filter. Retrieved J anuary, 2009, from:
Lizorkim, D. and Lisovsky, K., 2005. Implementation of the XML linking language XLink by functional methods.
Programming and Computer Software, Vol. 31, No. 1, pp. 34-46.
Lizorkim, D. and Lisovsky, K., 2003. SXML: an XML document as an S-expression. Russian Digital Libraries Journal,
Vol. 6, No. 2.
Lizorkim, D. and Lisovsky, K., 2003. XML Path Language (XPath) and its functional implementation SXPath. Russian
Digital Libraries Journal, Vol. 6, No. 4.
Lizorkim, D. and Lisovsky, K. XSLT and XLink and their implementation with functional techniques. Russian Digital
Libraries Journal, Vol. 6, No. 5.
Lizorkim, D., 2005. The query language to XML documents connected by XLink links. Programming and Computer
Software. Vol. 31, No. 3, pp. 133-148.
Lowe, D. and Wilde, E., 2001. Improving web linking using xlink. Open Publish, Sydney, AU.
May, W. and Malheiro, D., 2003. A logical, transparent model for querying linked XML documents. GI-Fachtagung
Datenbanksysteme fur Business, Technologie und Web.
May, W. and Malheiro, D., 2001. LoPiX: A systemfor XML data integration and manipulation. Proceedings of the 27th
VLDB Conference, Roma, IT.
May, W. and Malheiro, D., 2002. Querying linked XML document networks in the web. Proceedings of the 11th
International World Wide Web Conference, Honolulu, US.
May, W., Behrends, E. and Fritzen, O., 2008. Integrating and querying distributed XML data via XLink. Information
Systems, Vol. 33, No. 6, pp. 508-566.
Nentwich, C., Capra, L., Emmerich, W. and Finkelstein, A., 2000. Xlinkit: a consistency checking and smart link
generation service. ACM Transactions on Internet Technology, Vol. 2, No. 2, pp. 151-185.
Robie, J., 1999. XQL (XML Query Language). Retrieved J uly, 2010, from:
S-exp based XML parsing/query/conversion. Retrieved J anuary, 2009, fromSourceForge:
Silva, P.C. and Times, V. C., 2009. XPath+: A tool for linked XML documents navigation. XSym 2009 - Sixth
International XML Database Symposium at VLDB'09, Lyon, FR.
Silva, P.C., Aquino, I. J . S. and Times, V. C., 2008. A query language for navigation over links. XIV Simpsio Brasileiro
de Sistemas Multimdia e Web, Vitria, BR.
W3 Schools. XML Schema ComplexType Element. Retrieved J uly, 2010, from:
W3C. About W3C. Retrieved August, 2010, from:
W3C Recommendation. Namespaces in XML 1.0, Third Edition, 2009.
W3C Recommendation. XML Linking Language (XLink) Version 1.0, 2001.
W3C Recommendation. XML Path Language (XPath) Version 1.0, 1999.
W3C Recommendation. XML Schema part 1: Structures second edition, 2004.
W3C Recommendation. XQuery 1.0: an XML query language, 2007.
Walmsley, P., 2007. XQuery. O'Reilly, Sebastopol, US.
Wirth, N., 1996. Extended Backus-Naur Form(EBNF), ISO/IEC.
XBRL. Extensible Business Reporting Language (XBRL) 2.1. Recommendation, 2003.
XBRLAPI. A read-only API for XBRL. Retrieved J anuary, 2008, from XBRLAPI:
XLiP (XLink Processor). Users Guide, 2005. Retrieved J anuary, 2009, from:

ISBN: 978-972-8939-25-0 2010 IADIS
Takahiro Kawamura, I Shin, Hiroyuki Nakagawa, Yasuyuki Tahara and Akihiko Ohsuga
Graduate School of Information Systems, University of Electro-Communications
1-5-1 Chofugaoka, Chofu-shi, Tokyo 182-8585, Japan
Internet services using ontology in its background are rapidly increasing in the web. Those are sort of "lighter" ontologies
compared with ontologies which have been used in design and diagnosis. So, this paper proposes ONTOMO that is an
ontology building service specialized for lightweight ontologies. ONTOMO is also designed not for ontology experts, but
for users and/or developers of public web services. Here, we focused on ease of use, no installation, cooperative work
environment, and providing sample applications to help the users' ontology building. Especially, it has a function which
recommends instances and properties which belong to a certain ontology class in order to help the users to register them.
In the recommendation function, we developed our own Named Entity Extraction mechanism based on bootstrapping
method. In this paper, after the ONTOMO overview, we present the instance and property recommendation with
experimental evaluations.
Ontology, Named Entity Extraction, Web Service
Recently, web services using ontology has increased. Most of ontologies for those services are definitions of
metadata, and have simpler structures in comparison with conventional ontologies for design and diagnosis.
Faltings (2007) says that "is-a" relation makes up 80-90 % of all the relations in such lightweight ontologies.
However, ontology building tools proposed and developed so far are mainly for ontology experts, and few of
them are focusing on public developers and users of Internet services. Therefore, we have developed
ONTOMO, an ontology building service aiming at offering an environment in which those people are able to
easily make necessary ontologies for their purpose. So the main target of ONTOMO is the people who have
little expertise about what are an ontology, its schema, organization, and technical terms. Or, the people who
have some knowledge about the ontology, but no experiment of ontology building are also its targets.
ONTOMO has been trying to clear the following three hurdles that the target people may have in the
conventional ontology tools, which has been retrieved from interviews to non-ontology developers and
programmers around.
1. Necessary tool preparation and its usage are unclear, because of rich technical terms and functions.
2. It's hard to know what term should be input as classes, instances, and properties, because they have
never thought about things according to ontology own scheme.
3. Difficult to register a large amount of the terms (this point are relatively common regardless of the
non-expert or not).
To solve these problems, ONTOMO takes the following approaches.
A. Easy preparation for introduction
ONTOMO has no installation of any tools, and can be used by web browsers.
B. Improvement of usability (design of user interface)
It uses the public words without the technical terms, and narrows the functions down to ones only for
browse and edit the lightweight ontology. Also, it keeps visibility and operability even in the browser.
IADIS International Conference WWW/Internet 2010
C. Support of term registration
It would be almost impossible that a single person can think of every instances and properties in a large
domain. So, ONTOMO estimates candidates of instances and properties referring the registered terms, and
then recommends them to the user.
D. Keeping of motivation
To introduce the user's incentive, ONTOMO prepared sample applications to show what services can be
made by using ontologies. Also, it opens web service APIs to the ontologies to be used by any external
E. Support of collaboration work
Multiple accesses are possible for an ontology on ONTOMO, then the ontology can be built by a team.
The above three problems would not be solved by any single approach, and have 1-to-n relations to the
five approaches. Fig.1 shows the relations between the problems and approaches we have assumed.
Furthermore, their effect measurement for each problem would not be necessarily verified quantitatively,
because of the large dependency of user's feeling and sensitivity. Therefore, this paper firstly describes
overview of ONTOMO system, and then we would like to focus on the approach C. Support of term
registration with details of implementation and evaluation of the instance and property recommendation
function. Although the approach C is corresponding to the problem 3 and 2, we believe that an accuracy of
recommended terms can be regarded as a metric to measure the degree of the user support for these
1. Necessary tool
preparation and its usage
are unclear
2. hard to know what term
should be input as classes,
instances, and properties
3. Difficult to register a
large amount of the
Problem Approach
AEasy preparation for
BImprovement of usability
(design of user interface)
CSupport of term registration
DKeeping of motivation
ESupport of collaboration work
1. Necessary tool
preparation and its usage
are unclear
2. hard to know what term
should be input as classes,
instances, and properties
3. Difficult to register a
large amount of the
Problem Approach
AEasy preparation for
BImprovement of usability
(design of user interface)
CSupport of term registration
DKeeping of motivation
ESupport of collaboration work

Figure 1. Problems and approaches
The outline of this paper is as follows. We first describe the overview of ONTOMO with its interface and
functionality in section 2. Then section 3 shows the instance recommendation function and its evaluation,
where we compare the function with manual intervention to register the terms and show that it reduces the
users strain. Also, section 4 shows the properly recommendation and its evaluation. In section 5, we discuss
our limitation and improvement. Finally, we show the related works in section 6, and mention the future
issues in section 7.
ONTOMO logically consists of a front-end (Flash web site) and a back-end (web services) (Fig.2). The front-
end has an ontology edit function, the recommendation function, and two search applications as the samples.
Then, the back-end provides the APIs to access the ontologies built by ONTOMO so that other systems can
use the ontologies. We connected the front- and back-end by Asynchronous Flash, and realized high
responsibility and operability as a whole. This section describes the edit function, the search applications and
implementation issues of ONTOMO.
ISBN: 978-972-8939-25-0 2010 IADIS
Resource Layer
XML file
Client Layer
Web Browser
RPC Component
Edit func.
Server Layer
RPC Component
Ontology Processing Engine
Web Service
dation func.
Search Application
Resource Layer
XML file
Client Layer
Web Browser
RPC Component
Edit func.
Server Layer
RPC Component
Ontology Processing Engine
Web Service
dation func.
Search Application

Figure 2. ONTOMO architecture
2.1 Edit Function
instance class
Graph XML List Search
Selected node
Distance of nodes
Spring Graph pane
Tree pane
instance class
Graph XML List Search
Selected node
Distance of nodes
Spring Graph pane
Tree pane

Recommended instances
Recommended instances

Class list Property list Instance list Class list Property list Instance list

Figure 3. ONTOMO Interface
IADIS International Conference WWW/Internet 2010
The edit function is for the lightweight ontology, and has two roles: browsing major components of an
ontology such as the classes, instances and properties, and basic functions like new addition, edit, and delete
of them.
We provide three ways of the ontology browsing. The first one is Graph view to intuitively figure out the
ontology structure. Fig.3 top shows that Flash and Spring Graph which is an automatic graph allocation
library based on a spring model, visualize a hierarchical structure composed of the classes, subclasses and
instances. It can adjust, for example, a distance of an instance and a class with a focus on operability and
visibility. In addition, if the user moves a class, then instances of the class automatically follows and are re-
arranged with the pre-defined distance. Also, if the user double-clicks a class, the instances will hide, and
vice versa (unfortunately, we only have Japanese interface right now).
On the other hand, Graph view is unsuited to read through a list of the instances, and check the details of
their definitions. Therefore, we added List view and XML view. List view (Fig.3 bottom right) has three lists
for the classes, properties, and instances. XML view shows OWL (2004) data of the ontology as follows.

<rdf:RDF xmlns:mobilephone=
<rdfs:comment>MobilePhone OWL Ontology</rdfs:comment>
<owl:Class rdf:about="">
<owl:Class rdf:about=" phone"/>

Ontology editing is started with a popup menu which will appear by a right click on Graph view (the right
click is realized by Flash). If the user right-clicks a class or instance, the popup menu with "addition", "edit",
and "delete" will appear. Furthermore, the ontology built by ONTOMO is stored in DB in the back-end, and
can be exported in an OWL file, or accessed by the web services.
2.2 Search Application
ONTOMO also has a product search application, where the user can find products with key properties.
ONTOMO prepared the product ontologies of three domains in advance: mobile phones, digital cameras, and
media player like iPod. If the user searches any products with a common property like "SD card", the
products match that property will be found across the three domains. It's also possible to search with multiple
properties like hours of continuous operation and TV function, and then the ontologies of the mobile phones
and the media players will be searched.
Moreover, a blog search application shows the related blogs to the product found by the above using
Google blog search. The latest information about the product will be useful for the user and to update the
product ontology.
2.3 ONTOMO Implementation
ONTOMO is a 3-tier system (Fig.2), where a client layer provides a flexible user interface using Flex (2010).
Then, a server layer has an ontology processing engine with Jena (2009), and a resource layer stores the
ontology data in XML form with MySQL. Asynchronous call between the client layer and the server layer is
realized by Flex components like Flex Remote ObjectFlex Data Service. When the user operates the views
with Flash, the client layer sends a request to the server layer, and the Flex component calls corresponding
APIs. The result is returned in XML form for the graph or the list to the client layer, and then XML engine
transforms it to show on the Flash view.
ISBN: 978-972-8939-25-0 2010 IADIS
In the practical work of the instance addition, it would be difficult to register every term without any help.
Especially, consumer electronics, CD/DVDs, package software, etc. are increasing on a daily basis, so their
quick registration with a manual work would be impossible. Thus, we developed the function to recommend
candidates for the instances based on records of the user's registration. This function estimates the instances
which belong to the same class with the registered instances as seeds using Named Entity Extraction (NEE).
But, the user finally selects and adds some of them as new correct instances. For example, if the user registers
"Nissan" and "Honda" as the instances, and added "Toyota" as the third instance, then the function takes the
latest three instances {"Nissan", "Honda", "Toyota"} as the seeds for the NEE, and recommends other
candidates for the instances to the user after exclusion of the already-registered ones (Fig.3 bottom left).
3.1 Named Entity Extraction with Bootstrapping
There already exist free Named Entity Extraction services such as Google Sets (2007) and SEAL (2008) on
the web. However, our investigation found that those have maintenance difficulty for the quick registration of
the latest information. Therefore, we developed our own Named Entity Extraction engine using bootstrapping
method (Brin 1998). The bootstrapping firstly generates patterns from documents using a small number of
the seed terms, and extracts other terms from the documents using the patterns. Then, it generates new
patterns using the extracted terms as the new seeds again. This repeated process can extract a large number of
terms from a few seeds. The following describes our bootstrapping process with four steps.
(1) Selection of seeds
We first need some seeds to execute the NEE. Selection of the seeds greatly effects on the extracted terms.
We prepare a seed list which includes at least three instances in advance. At the first time, it's random
selection and the bootstrapping is executed, and the extracted terms are added to the seed list. At the second
time, we take also randomly two old instances used before and new one extracted as the seeds, because newly
extracted instances may not be necessarily correct. So we take old two and new one after the second time for
extending the terms and keeping the accuracy.
(2) Collection of Web pages
We collect the top 100 web pages using search engines like Google and Yahoo!, where the query is a
combination of three instances.
(3) Pattern generation
Then, we find HTML tags which surround all three seeds from collected web pages. For example, if the
seeds are {"Nissan", "Honda", "Toyota"}, the surrounding tag is as follows.


Here <td> tag surrounds all three tags, then <td>term</td> is taken as a pattern. The reason we find the
tag surrounding all three seeds is to keep the accuracy. The stricter the condition is, the higher the accuracy
becomes. Our preliminary experiences for some kinds of product showed that in case of two seeds the huge
numbers of irrelevant terms were extracted. But, more than four seeds greatly reduced the extracted terms.
Therefore, this number of seed is our experimental setting, and should be revised at least for other domain.
(4) Extraction of terms
Finallywe extract other terms from the same document using the generated patterns. Although web
pages have several styles according to their authors, it seems that the same style (pattern) would be used at
least within the same document. In the example of (3), we can take "Matsuda" and "Suzuki" from this
document, and add them to the seed list.
However, we still have a possibility to extract the irrelevant terms in the above process. So we set
threshold for the number of patterns which can be generated in a document. In our experience, we have been
able to extract the terms most effectively in case that the number is more than two. Therefore, if the
generated patterns are more than two kinds, we add them to the list. If not, they are discarded.
IADIS International Conference WWW/Internet 2010
3.2 Instance Recommendation with named Entity Extraction
The set of terms extracted by the above process is not necessarily the same as the one the user expects as the
instances for a certain class. Naive recommendation may put a heavy strain on the user for proper selection of
a large number of terms. This paper, therefore, proposes an adaptation filter, which combines the NEE and
the hierarchical structure of the ontology to raise the accuracy of extraction.
If the intended set Si of the user is equal to the extracted set Se, then all members of Se are registered to
the list. This is an ideal case. However, if not, we have two cases.
3.2.1 Improvement of Recall
If only a few terms of Si are included in Se (Fig.4 left), recall will be low because of few correct terms in Se.
So we add some terms of Si, which are not included in Se, and try to extend Se to Se' (experiment 1).
Need to add
Extended set
Need to add
Extended set

Recommended terms
Need to add
Need to delete
Recommended terms
Recommended terms
Need to add
Need to delete

Figure 4. |SeSi|<<|Si| and |SeSi|<<|Se|
3.2.2 Improvement of Precision
If a number of the irrelevant terms are included in Se (Fig.4 right), precision will be low because of many
incorrect terms in Se. So we apply the adaptation filter to eliminate the irrelevant terms Sx, and raise the
precision (experiment 2). The following adaptation filter is a mechanism to eliminate the terms which might
be irrelevant from the instance set returned by the NEE.
First, eliminate overlapping terms comparing with the already-registered instances which belong to
the same class as one of the seed instances.
Next, execute the NEE again with the new seeds which belong to a different class than one of the
original seeds. Then, eliminate overlapping terms comparing with the newly retrieved set.
For example, Fig.4 right shows a case that we have a car ontology like Fig.5, and want to add some new
instances to Japan car class (Si). The result set Se has the irrelevant terms like "Daikin", "Ford", "GM",
"Sony", and the precision is low. Then, the adaptation filter first eliminate the overlapping terms comparing
with the already-registered instances, which belong to the same class (Japan car) as one of the seed {"Nissan",
"Honda", "Toyota"}. Secondly, it executes the NEE with the seeds from another class, that is, {"GM", "Ford",
"Chrysler"} of US car in Fig.5. The result set Sx has high possibility to include US car instances, so that the
overlapping terms in Se with Sx are eliminated. But, we should note here that if Sx has Japan car instances,
then those are also eliminated and may lower the recall.

Toyota Nissan Honda
GM Chrysler Ford
Toyota Nissan Honda
GM Chrysler Ford

Figure 5. Example of car ontology
ISBN: 978-972-8939-25-0 2010 IADIS
3.3 Evaluation
As mentioned above, the instance recommendation is mainly to stay up-to-date in a domain with dynamic
nature. However, we took a relatively static domain for this experiment in order to accurately measure the
precision and the recall. Here, we focused a world ranking of car manufacturers (2007), and took the top 50
car manufacturers (and their spin-off brands) in total 76 as the correct set in the experiment 1, excluding
china and OTHER manufacturers. It's because the current china manufacturers, that is, the correct set seem to
have been changed from the above data in 2007, and have fluctuation of description and many spin-off
brands that we could not investigate. In the experiment 2, total 16 of Japan manufacturers and their brands
are taken as the correct set. Also, in the experiments we regarded other descriptions than ones in the correct
set as incorrect. In the original use, the instance recommendation leaves the final selection for the addition of
the extracted terms to the ontology to the user. But, in the experiment the function automatically adds every
extracted term expect for the overlaps in order to confirm the accuracy. In the following sections, we evaluate
how we can raise the precision and the recall using the NEE and the adaptation filter.

Recall = Registered correct instances / All the correct instances
Precision = Registered correct instances / All the registered instances (including irrelevant ones)
3.3.1 Experiment 1 (Named Entity Extraction)
In the experiment 1, we evaluated how we can effectively increase the instances in contrast to the number of
operations using the NEE for the 76 world car manufacturers. As the first seeds we registered {"Nissan",
"Honda"} to Car class, then take the next instance "Toyota" according to the listed order of the correct set
except for the overlaps. So, the NEE is executed with the latest registered seeds {"Nissan", "Honda",
"Toyota"}, and gets the candidates for the instances Se. Then, it registers all of them to the class except for
the overlaps. When all the correct instances are registered by repeating the above, this experiment is
terminated. Evaluation metrics are as follows.

Recall = (3 + ) / 76
Precision = (3 + ) / (3 + + irrelevant instances), where is the registered correct instances

The result is shown in Fig.6. In terms of the recall, because simple input means to manually register the
correct instance one by one, the recall moved up in proportion to the number of operation, and achieved
100 % at the 73rd time. On the other hand, because the NEE extracted a number of instance candidates at the
early stage, the recall moved up quickly, and achieved 71.1 % at the second time. After that, however, the
extracted terms began to overlap with the registered ones and the move became slowly, then finally it
achieved 100 % at the 24th time. As a consequence, we confirmed the NEE increases the instances
effectively with a small number of operations.
In terms of the precision, it remained 60.3 % even at the last time due to the irrelevant terms extracted in
Named Entity Extraction, although the simple input always kept 100 %.
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Tim e Tim e Tim e Tim e
Sim ple Input
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70
Tim e Tim e Tim e Tim e
Sim ple Input

Figure 6. Result of experiment 1

IADIS International Conference WWW/Internet 2010
3.3.2 Experiment 2 (Adaptation Filter)
In the experiment 2, we evaluated how we can keep the precision using the adaptation filter for the 16
Japanese car manufacturers. Because the adaptation filter uses the hierarchical structure of ontology as
mentioned above, we assumed here that Japan car is a subclass of Car and US car is also a subclass of Car in
the same level as Japan car like Fig.5. As the first seeds we registered {"Nissan", "Honda"} and {"GM",
"Ford", "Chrysler"} to Japan and US car respectively. Then, we repeated the procedure of the experiment 1
with the adaptation filter to eliminate the instances which seems to belong to a different class (US car).
Evaluation metrics are as follows.

Recall = (3 + ) / 16
Precision = (3 + ) / (3 + + irrelevant instances), where is the registered correct instances

The result is shown in Fig.7 (note that 0 time of the precision is omitted here). In terms of the precision,
the NEE was 19.5 % at the 7th time due to US and European manufacturers included in the instances. But,
the NEE with the adaptation filter got 26.1 %, and the gap was extending over time. As a consequence, we
confirmed the adaptation filter has effect to raise the precision.
However, in terms of the recall, the adaptation filter remained 66.7 % at the 7th time, although Named
Entity Extraction achieved 100 % at that time. It's because some Japan car instances were also eliminated
together with the irrelevant terms.
0 1 2 3 4 5 6 7
Tim e Tim e Tim e Tim e
NEE NEE w/ Filter
1 2 3 4 5 6 7
Tim e Tim e Tim e Tim e
NEE NEE w/ Filter

Figure 7. Result of experiment 2
4.1 Property Recommendation with Named Entity Extraction
As well as for the instances, it would be difficult to register every property without any help. Property
recommendation collects the properties using the NEE for the bootstrapping as well. But, in case of the
property, we use three properties belonging to a class and an instance of the class as the query to collect the
web pages. This is because adding the instances would be useful to collect the related web pages like product
specification pages. The seeds for the NEE also include three properties and an instance. For example, if the
user puts "optical zoom", "digital zoom" and "pixels" to a camera class, then the property recommendation
takes those three properties and an instances of the camera, and extracts other properties like "focal length",
"sensitivity", and "shutter speed" to recommend for the user. After that, the seeds are changed to two old
properties, one new property, and one new instance. In essence, the NEE using the bootstrapping is to extract
terms written in the same style. So, we can expect that the seeds in different levels, that is, the property and
the instance do not work well. But, most of the product specification pages have a product name at the
beginning, then its specs (properties) later, so that we considered this seed combination would not affect so
much for the property extraction.
ISBN: 978-972-8939-25-0 2010 IADIS
4.2 Evaluation
A property belongs not only to a certain product domain, but also to multiple ones. For example, "screen
size" is for a digital camera, and also for a laptop PC. Therefore, it is difficult to explicitly classify the
properties to each class unlike the instances, and the adaptation filter has a possibility to greatly reduce the
recall. So, the experiment uses the NEE without the adaptation filter. We selected the target product domains
(classes) which relatively have many properties: camera, lens, TV, PC, LCD, Blu-ray player, and made the
correct properties set. A flow of the experiment in case of the camera is as follows. We first register the
properties {"optical zoom", "digital zoom", "pixels"} as the seeds to the camera class in advance, and get an
instance of the camera using the instances recommendation. Then, we execute the NEE with the seeds {a
camera instance, "optical zoom", "digital zoom", "pixels"}, and extract the candidates of the properties.
The result is shown in Fig.8. We confirmed the recall achieved 70-80 %, and almost all the properties of
the class were extracted. However, the precision remained 50-60 % on the whole. Although it's ideal to
extract all the properties from pages such as the product specification of the manufacturers, this experiment
has not selected the page to be extracted. As a consequence, it extracted a number of the irrelevant terms
from the pages like explanation of the property meanings, and they lowered the precision.


Figure 8. Result of property recommendation
In terms of C. Support of term registration in section 1, we presented that the instance and property
recommendation reduces the user's strain. The instance recommendation also introduced the adaptation filter
due to insufficient precision of the Named Entity Extraction. As the result, it raised the precision though it
lowers the recall. However, the reason we rather emphasize the precision is that it would be more difficult for
the target users of ONTOMO who are not experts of the ontology to pick the correct terms from a large set
with many irrelevant terms (Of course, it depends on the domain knowledge). On the other hand, in case of
the low recall the user needs only to repeat the process of the recommendation, which would be a relatively
simple task (if the precision keeps high). Furthermore, the property recommendation could not use the
adaptation filter, so that to improve the precision we are now considering introduction of a process to select
the pages to be extracted by looking at hit count of search, and so forth.
However, the ontology which can use this recommendation function is limited to ones, where a class is
corresponding to the already-existing category like a product domain. This is a kind of limitation of the NEE
from the web, which means that it can not collect the terms of a classification unless someone classified it in
advance. Moreover, the current procedure of the instance recommendation can not classify the classes which
are not determined by just three terms. For example, the instances {"Tokyo", "Shinagawa", "Shinjuku"} are
for Tokyo City class, and also for Train Station class, and then those are not determined automatically. But,
as the property recommendation we can add a typical keyword like "City" or "Station" when searching the
pages, and specify the classes to some extent. In the future, we will discuss this improvement with our pattern
extraction method.
In terms of other approaches: A. Easy preparation for the introduction and B. Improvement of usability
(design of user interface), as mentioned in section 1 we skipped the tool install by using the web browser, and
IADIS International Conference WWW/Internet 2010
realized operability of a desktop application using Flex. Also, Spring Graph realized visibility of the whole
ontology and the details of the definitions. In terms of D. Keeping of motivation, we are trying to motivate
the user by providing sample applications and open web APIs, and for E. Support of collaboration work an
exclusive access control of ontology DB enabled multiple developers to build an ontology at once. For those
approaches, however, there are some other ways like improvement of documents, innovative user interface,
and point system to registration, etc. In the future, we will compare several ways and discuss their effect
This section shows comparisons with other ontology building tools and Named Entity Extraction methods.
As the famous ontology building tools, we compare with Protg, Hozo, and KiWi. Protg (2010) is the
most popular ontology editor mainly used by ontology researchers for more than ten years, and has several
advanced features like inference. Also, it opens plug-in specification, and now has 72 plug-ins in its official
site such as data import from RDB, term extractions by text mining, etc. Hozo (2010) is a tool which has a
unique feature to handle Role concept. It also has a distributed development environment which can keep
data consistency by checking difference of ontologies edited by multiple users. However, those are the
heavyweight ontology tools mainly for the ontology experts, thus different from ONTOMO in its target user
and problems to be solved. In the future, we will consult them with their ontology visualization plug-in and
mechanism to keep the consistency of data, and so forth.
On the other hand, KiWi (Knowledge in A Wiki) (2010) focused on the lightweight ontology, and
extended Wiki by semantic web technology. It enables the user to edit its content through the same interface
as Wiki, so the user can register the instances and properties without any special knowledge of ontology. It's
a different way than ONTOMO, but for the same purpose in easy introduction for non-expert, improvement
of usability, and collaborative work. In the future, ONTOMO would also like to take Wiki-based change
history and difference management.
They all do not have any recommendation function for the instances and the properties, so that we could
also distinguish ONTOMO by this function. But, we next compare this function with DIPRE and Snowball,
which are the famous Named Entity Extraction methods for structured documents. DIPRE (Brin 1998) is a
pioneer of the bootstrapping from the web pages. First, it prepares a list of pairs of an author name and a
book title, and searches them by a search engine. Then, if it finds a regular expression like "URL, prefix
string, book title, middle string, author name, suffix string" from the result, it adds the pairs of the author
name and the book title which match the regular expression. Snowball (Agichtein 2000) is an extension of
DIPRE, and tried to extract named entities in comparison with the string pairs of DIPRE. In contrast, we tried
to extract the terms (instance, property) which belong to a certain set, and has a feature that the adaptation
filter eliminates the irrelevant terms using the hierarchical structure of the ontology, in addition to setting the
threshold to the number of patterns based on the experiment, and so forth. In the future, we consider any
performance evaluation with the above-mentioned Google Sets and SEAL managing to make the same
condition for them.
This paper presented development of an ontology building service, ONTOMO. In addition to the above
future issues, we will consider an automated construction of ontology itself in the future. Web pages which
have relatively high credibility of its content like Wikipedia would be a good source of the ontology. There
are already some researches to build the ontology from part of Wikipedia content, but a point in the future
would be how we merge the information from a variety of sources accurately. We will continue to offer the
ontology tool for the public developers and users of the Internet services incorporating a diverse range of
their opinions.

ISBN: 978-972-8939-25-0 2010 IADIS
Adobe Flex, <>.
B. Faltings, V. Schickel-Zuber, 2007. Oss: A semantic similarity function based on hierarchical ontologies. Proceedings
of 20th Int. Joint Conf. on Artifical Intelligence (IJCAI 2007), pp.551-556.
Boo!Wa! - A List Extractor for Many Languages (former Seal), <>.
E. Agichtein, L. Gravano, 2000. Snowball: Extracting relations from large plain-text collections. Proceedings of 5th
ACM/IEEE-CS Joint Conf. on Digital Libraries (JCDL 2000), pp.85-94.
Google Sets, <>.
Hozo, < >.
Jena - A Semantic Web Framework for Java, <>.
KiWi - Knowledge In A Wiki, <>.
OWL Web Ontology Language, <>.
S. Brin, 1998. Extracting patterns and relations from the world wide web. WebDB Workshop at 6th International
Conference on Extended Database Technology, pp.172-183.
Spring Graph, <>.
The Protg Ontology Editor and Knowledge Acquisition System, <>.
World Ranking of Manufacturers Year 2007, <>.
IADIS International Conference WWW/Internet 2010
Claudia Iacob and Li Zhu
Dipartimento di Informatica e Comunicazione, Universit degli Studi di Milano
Via Comelico 39, 20139 Milano (Italy)
Web systems provide solutions for real world problems from various domains. Defining a problem space and
materializing its solution within a web space brings a set of challenges that come from both the diversity of end users
facing the problem and the diversity of technologies available today. This paper addresses the design of Web systems
seen as web spaces which provide their end users with tools tailored to their domain, their role in the domain, their
culture and the platform they are using. These tools help end users develop their exploration of the problem at hand
without being lost in the space. We propose an abstraction model able to support the design of localized web systems and
which preserves software quality measures like maintainability, reuse and consistency. The method for applying this
model together with a concrete example of its application is presented.
Abstraction, languages, software design, Web.
In the Web 2.0, there is an increasing need of allowing end users i.e. people who are not computer science
experts, but are supported by software systems in performing their everyday work activities (Brancheau,
1993) to access applications and information in a seamless way, using different platforms (Lin, 2008) or
acting different roles (Carrara, 2002). Many of these end users often develop creative activities, such as
design (Xu, 2009), medical diagnosis (Costabile, 2007) or artistic creation (Treadaway, 2009). The goals of
these activities are not sharply defined at the beginning of each activity, but emerge progressively by the
exploration of the problem space. The end users develop their creative processes as unfolding processes, in
which their problem space is progressively differentiated to create a complex solution (Borchers, 2001).
End users therefore need to access a web space, which is a space of opportunities (Alexander, 1977) where
they find tools which allow them to develop their exploration without being lost in this space and converging
to their final results. In their activities, end users of different cultures and working in different domains often
face a common problem and collaborate to reach its solutions. They need to work in teams, share tools as
well as knowledge and wisdom about the problems solutions; moreover, they need to access all resources
according to different styles of interaction, which respect different cultural conventions. Hence, they use
different communication and reasoning tools in their collaboration, giving rise to a set of issues and
difficulties to be faced.
Based on the study of previously designed web systems addressing various problem domains such as
medicine (Costabile 2006), engineering (Valtolina, 2009), geology (Carrara, 2000) - we argue that end users
need to be supported in their daily work activities by being allowed to interact with tools tailored to their
domain, role, culture and platform in use. These tools must reflect the problem domain and need to be
materialized and managed in the Web. Moreover, as experts in their own domains, end users should be
supported to freely express their knowledge of the domain and intentions by using the tools they are
accustomed with. Therefore, we aim to design web interactive systems (WIS) which support end users in
their work by: i). allowing them to reason in their own system of signs and to exploit their skills and
knowledge, ii). mitigating - on the side of the users - the process of changing the flow of reasoning from one
problem domain to another and iii). allowing the materialization of a WIS on diverse digital platforms.
ISBN: 978-972-8939-25-0 2010 IADIS
Research in modeling WISs is being done on several levels. Of interest to our work is the modeling of
WISs in ways that allow the shaping of the system to the user and the platform in use while preserving the
quality standards of software design. For example, in (Lin, 2008) the materialization of a web system is
abstracted by a set of design patterns. These patterns can be applied as a starting point of the design of the
specific system, considering that they provide a solution to a common and usually broadly met problem. The
major disadvantage of this kind of approach is that the collection of generic design patterns is limited and
may not address all the problems of design. In (Calvary, 2003), a framework Cameleon Reference
Framework for describing multi-target user interfaces is presented. The framework follows an object-
oriented approach, modeling at an abstract level the domain of the problem in terms of objects and the
relationships between them. Moreover, the behavior of the system is modeled at the level of task comprising
the description of a set of inter-related tasks. UsiXML (Molina, 2005) is a language used to describe abstract
user interfaces which are independent from the hardware and software platform in use, the physical
environment where the interaction takes place and the users abilities.
The model presented by this paper follows a different approach in that it models the dynamics of each
individual entity in the system separately. The behavior of the overall system is never modeled as a whole,
but decided by the end user during her/his interaction with the system on the basis of the possible actions s/he
can perform on each entity. Moreover, the model allows the localization of a WIS on four levels: problem
domain, end users role and culture and the platform in use.
The section proposes an abstraction model to be used for designing WISs localized to the problem domain
the system is a solution for, to the end users profiles (in terms of their role and culture) and to the platform in
use. The model is defined by a hierarchy of levels of abstraction (Figure 1). Each level is described by one or
more XML-based languages which make possible the definition and localization of the virtual systems to the
end users profiles (domain, role, culture and platform). The hierarchy of levels is organized using a top-
down approach, following the description of the web system from the most abstract level of definition to its
concrete instantiation.

Figure 1. The hierarchy of abstraction levels of the design model proposed
1). The top level the meta-model level describes the rules for defining domain-specific languages able
to define WISs, seen as web spaces in which tools to support domain knowledge-related activities are made
available to the end users. This level prescribes how to define the common structure of WISs, describing
them as web spaces and abstracting from their end users roles and cultures and the platforms in use. The
meta-model level is defined by a meta-language Interaction Multimodal Markup Language (IM
L). IM
L is
used for defining domain-specific object languages at the domain level.
2). The domain level localizes the WISs to a specific domain. In each domain, end users may have
different roles. At the domain level, the common characteristics of the web space defined for end users
IADIS International Conference WWW/Internet 2010
having different roles in a specific domain are described. These common characteristics are related to the
knowledge-related activities specific to the domain and to the structure of the web space associated to the
WIS. From the definition of IM
L, domain-specific object languages called G-IM
L - can be derived. Each
such language is tailored to be highly productive in defining web spaces for a specific problem domain.
Therefore, for each application domain, one G-IM
L language can be derived. A set of documents written in
L describes a web space for a particular domain and a specific role in the domain
3). The configuration level comprises two sublevels: the culture localization and platform materialization
sublevels. End users having the same role in the same domain may belong to different cultures and may use
different platforms to interact with the WIS. The configuration level localizes the web space defined at the
domain level to the culture of the end user and to the platform s/he is using. The configuration level is
supported by the definition of two languages Localization Markup Language (LML) that addresses the
culture and Template Language (TL) which addresses the platform.
4). The instance level comprises instances of a WIS defined at the domain and configuration level and
materialized on a specific platform. At this level, the end user can access the WIS, explore it as a web space
and interact with the tools offered within it to perform the activities of interest to him/her.
5). The system level comprises the family of WISs addressing the same domain, but diverse end users
(with different roles and cultures) and platforms.
2.1 Meta-model Level
L is a XML-based meta-language whose specification allows the derivation of domain-specific languages
which describe WISs as web spaces localized to a domain. Within a web space, the end user can live some
experiences and make events happen (Borchers, 2001) by performing knowledge-related activities which
allow the externalization of implicit knowledge. IM
L describes the rules for defining the characteristics and
the refined structure of web spaces (Alexander, 1977). IM
L meta-language prescribes how to define domain-
specific languages which allow the description of WISs abstracting from the end users roles, cultures and
platforms in use. In this way, IM
L defines an abstraction model which allows the design and the
implementation of personalized instances of WISs which allow their localization to a domain, and end users
role and culture and a platform. To express the model, we unify and group together the common
characteristics of different knowledge activities based on a knowledge base which must be created,
maintained and updated. The activity of managing the knowledge base is the annotation, which allows end
users to update it by adding comments next to data in order to highlight this datas meaning (Costabile,
2007). Therefore, annotation supports and enhances the communication, as well as the knowledge
L defines a web space as superimposing a set of specialized subspaces: infoSpace, workspace,
operatorSpace, messageSpace and transparentSpace. An entity (Costabile, 2007) e
in this case, a
specialized subspace - is superimposed to another entity e
in this case, the spaces - when e
belongs to e1
and every action on both the entities is only handled by e
and does not reach e
(Bottoni, 1997). The
classification of spaces is made on the basis of the pattern of events made available by each space. An
infoSpace allows the end user to get directions on how to navigate through the space and understand it.
Examples of entities superimposed to an infoSpace are labels and titleBars. A workSpace is the working area
of the system, in which the end user develops activities of interest to her/him, including annotating the
knowledge base. As example, the workSpace in an application like Google Maps is the entity on the screen
containing the map. The operatorSpace permits the end user to interact with the system. A toolbar is a
concrete example of operatorSpace, met in a large variety of web systems. The messageSpace allows the end
user to receive feedback from the system s/he interacts with. As concrete example, a warning message is
superimposed to a messageSpace.
Each type of space superimposes entities possibly other spaces which define it and allow the end user
to perform activities within it. Each entity is defined by a set of characteristics, N = {ID, DESC, T, D, IS, C,
S} where:
a). ID uniquely identifies the entity.
b). DESC shortly describes the entity.
c). T denotes the type of the entity.
ISBN: 978-972-8939-25-0 2010 IADIS
d). D denotes the dynamics of the entity. The dynamics of an entity describes its behavior in response to
the end users actions. This behavior is identified by a set of states an entity could have and a set of
interaction steps through which the entity moves from one state to another.
D := <State>+ <Step>*
Any state must be identifiable and may express at least one property like gravity, emotion, complexity.
State := <ID> <PROPERTY>+; PROPERTY = {Emotion, Complexity, Gravity}
For example, an operator could express an alarming emotion providing that some condition is met.
An interaction step specifies the fact that an entity which is in a state, statePre, moves into a new state,
statePost, under an action from the end user and a possible additional condition. The actions that an end user
can perform on an entity are classified as Selection, Indication and Activation. As example, moving the
mouse over a button triggers an Indication action.
Step := <ACTION> <statePre> <cond> <statePost> <Effect>*
ACTION = {Selection, Indication, Activation}
statePre, statePost are of type State
cond is the condition under which the transition (statePre, ACTION) -> statePost (1) occurs
Moreover, this transition may trigger several effects on other entities.
Effect := <ID> < effectEntity> <newState>
ID identifies the effect uniquely
effectEntity is the ID of an entity defined in the system which is affected by the transition (1)
newState is the ID of a defined state for the entity effectEntity. Following the transition (1), effectEntity
will change its state into state newState
e). IS is the representation of the initial state of the entity. The initial state follows the definition of a
generic state, but identifies the representation of the entity at the start-up of the system.
f). C is the context of the entity, that is the set of entities to which this entity can be superimposed.
g). S is the structure of the entity, which is the set of entities which can be superimposed to it.
The tree structure of the IM
L language makes it feasible for the overall definition of IM
L to be specified
by a XML Schema. Hence, each entity: i). is being defined by the set N = {ID, DESC, T, D, IS, C, S} and ii).
is identified by its position in the XML Schema.
2.2 Model Level
Each WIS is associated with a domain-specific web space. From the definition of the IM
L meta-language, a
set of domain-specific object languages called G-IM
L languages can be derived. Each G- IM
L language
allows the description of web spaces designed for a particular domain. A G-IM
L language localizes a WIS
to a specific domain, describing the knowledge activities to be performed in that domain by end users playing
various roles in different cultural contexts. A set of documents written in G-IM
L defines a web space for a
specific role in a particular domain, describing the domain-specific activities allowed to end users with a
particular role in the domain. Moreover, it abstracts this definition from cultural details and from
materialization specific technicalities.
Respecting the definition rules provided by the IM
L meta-language, any G-IM
L language defines the
web space, a set (or all) of the specialized subspaces and a set (or all) of the entities superimposed to them in
terms of:
a). ID: Any G-IM
L must define a unique ID for each entity. The rules for defining the ID are domain
dependent, so each domain specific object language defines them independently. For example, a specific G-
L language may define an ID as a string of characters containing at least one digit.
b). DESC: An entity defined in a G-IM
L language may be described by a significant name in the domain
of the language.
c). T: Entities defined in G-IM
L languages may have a type attribute. Each domain defines its specific
types for each entity defined. For example, the admissible types for a workshopSpace may be ordered and
unordered: T = {ordered, unordered}. For an ordered space, one would assume that the order of the entities
superimposed to the workshopSpace matters, while for an unordered workshopSpace this order does not
d). D: Conform to the IM
L definition of a state, each domain defines the structure of the states of an
entity within the domain together with the possible actions an end user can perform on the entities defined for
IADIS International Conference WWW/Internet 2010
the domain. For example, a particular domain-specific G-IM
L language may define a state as being
identifiable and expressing an emotion.
State := <ID> <Emotion>; Emotion = {appreciation, surprise, disappointment, sense of danger}
Moreover, each G-IM
L language defines the possible actions specific to the domain associated to the
L language.
e). IS: The initial state is a particular state which follows the same definition as a generic state. Any entity
defined in a domain specific G-IM
L must have an initial state associated to it.
f). C and S: The definition of a G-IM
L is given by an XML Schema, hence the context and the structure
for each entity is embedded within the tree structure of this schema. The entities described by the children of
a node represent the structure, S, of the entity described by the node. The context, C, of an entity is
represented by the entities described by the nodes in the path from the root node of the tree to the node
representing the entity.
The entities defined by a G-IM
L language provide means of defining abstractly web spaces tailored to a
specific domain and providing the end user having a specific role only with the tools required for
performing the specific domain related knowledge activities.
2.3 Configuration Level
Tailoring WISs to the end user and the platform in use requires the localization of the domain-specific web
space (associated to the system) to the culture of a certain end user, and to the platform s/he uses.
Consequently, we define two mark-up languages Localization Markup Language and Template Language
able to: i). enhance the abstraction model defined by IM
L ii). progressively shape an instance of the WIS
abstractly described by G-IM
L documents (and defined for a domain and a role within the domain) to a
specific end user by describing localization and materialization details about the instance.
Localization Markup Language (LML)
LML localizes a web space to a specific culture. The goal of LML is to describe the culture-related
properties of the domain-specific web space associated with an instance of a WIS. These properties are: color
representation, shapes, text alignment, language, and topology (Wright, 1997). LML is a mark-up language,
its elements being documents respecting the LML definition given by a XML Schema (Barricelli, 2009b).
Each targeted culture is associated with a LML document. Text alignment is a culture related property which
specifies the order in which text is displayed for one particular entity or for the overall web space. The
possible values for this property are: left-to-right or right-to-left.
Color representation is a culture-dependent property which is described in LML by functions associating
each G-IM
L defined value for each property of a state to a color embedding the meaning of the propertys
value. For example, for a state transmitting an emotion, each possible emotion transmitted is associated to a
color which embeds the meaning of the specific emotion in a particular culture. Table 2 provides a mapping
between the color associated to emotions for the Italian and the Japanese cultures.
Table 1. Italian and Japanese color-emotion mapping
Emotion Italian Japanese
sense of danger

The instance of the system for the Italian culture will use yellow for representing the states expressing
appreciation, while the instance for the Japanese culture will use green to represent the same states.
Language is addressed in LML by a language dictionary repository which assigns to each G-IM
L defined
entity the translation of the text associated with it in the language of the culture targeted. Topology specifies
the order in which entities will be represented for a specific culture. The shapes property associates each
entity with a set of shapes that may be used for representing it for a specific culture.
Template Language (TL)
TL localizes a web space to a specific platform. TL provides means to describe the way each instance of
the WIS will be materialized on the platform in use. TL describes the platform specific properties of the
ISBN: 978-972-8939-25-0 2010 IADIS
instances of WISs and the way their structure is affected by these properties. Examples of materialization
properties are relocatable geometry, and graphical properties (such as the lines thickness). As example, an
instance of a WIS materialized on a laptop will address the specificities of the display of the laptop, while an
instance of the same system designed for a mobile device will respect different properties of materialization
accordingly to the platforms specification (memory, input/output devices, display size, resolution and so on).
The elements of TL are documents which respect the schema definition of TL (Barricelli, 2009b).
A TL document describes a set of templates to be applied for a specific platform. Each template
associates a G-IM
L defined state of an entity with the parameterized snippet of SVG or HTML code to be
used for its materialization on a specific platform. As example, describing the materialization of an
information type entity, its initial state may be associated to a snippet of code which will describe a label and
will replace the text and the color of the label with parameters whose values will be decided at run-time
based on the culture of the end user.
This section describes a step-wise design method for reaching concrete instances of WISs applying the
aforementioned model. The method follows a top-down approach and progressively shapes personalized
instances of WISs. The sequence of steps to be followed during the instantiation of the system has as starting
point the existing definition of the IM
L meta-language (Meta-Model Level in Figure 2), based on which
domain-specific G-IM
L languages can be defined. A set of G-IM
L documents (Domain Model Level in
Figure 2) abstractly describes an instance of a WIS, seen as a web space localized to a domain and addressing
end users having a specific role in the context. The G-IM
L definition of the domain-specific web space is
being augmented at the configuration level by the culture localization (Culture Configuration sublevel in
Figure 2) and the platform materialization (Platform Configuration sublevel in Figure 2). The culture
localization phase adds to the web space the culture-related properties described by the LML documents,
while the platform materialization phase binds it to a specific platform, by applying to it the properties
described by the TL documents.

Figure 2. The flow of reasoning in reaching families of interactive virtual systems from IM
Culture localization and platform materialization properties are merged with the abstract description
defined by the G-IM
L documents, reaching an instance of the WIS (Instance Level in Figure 2). Any
instance of a WIS addresses a specific domain, an end users culture and role in the context and a platform.
The set of all instances of a WIS defined for a specific domain and for end users having different roles,
cultures and using different platform form a family of system (System Level in Figure 2).
The flow of execution for reaching a particular instance of a WIS starts with checking the profiles of the
end user who accesses an instance of the WIS, in terms of her/his role in the context, her/his cultures, and the
platform in use. The G-IM
L documents are then loaded together with the culture localization (LML) and
IADIS International Conference WWW/Internet 2010
platform materialization (TL) documents according to the end users profiles. The LML and the TL
documents become the input for a XSLT processor. The XSLT processor merges the LML and TL
documents, creating a localized template for the instance. The localized template is further on merged with
the G-IM
L documents. The result of the XSLT processor is a full tree representing the initial state of the
entire instance of the WIS, which is then materialized on the screen and ready to react to the end users
actions. The end users actions are captured as events and are treated according to the dynamics rules
embedded in the G-IM
L documents.
3.1 A Concrete Example
The design method proposed by this paper follows an iterative approach in that it allows, at each step of its
definition, the design and the implementation of concrete examples of WISs. The example presented in this
subsection the Valchiavenna Portal - is one iteration of the method (Barricelli, 2009). The portal addresses
the tourists and domain experts interested in a valley situated in the northern part of Italy, called
The design of the portal initiates with deriving from IM
L a G-IM
L object language addressing the
touristic domain. This domain-specific language provides means of defining touristic web systems which
allow viewing and annotating points of interest on a map of the Valchiavenna region. The abstract
characteristics and the structure of these spaces are described by two sets of documents written in the
touristic-specific G-IM
L object language: i). one set describes the touristic web space for tourists and ii). the
second set describes the touristic web space for domain experts (such as historians, geologists). The common
activity for both types of end users of the portal is the annotation they are all allowed to annotate points on
the map. However, for each role, different annotation means are described. Tourists are allowed to associate
points on the map with visual links (emoticons) to the personal comments on their visit of the place. On the
other hand, domain experts are provided with tools to associate points on the map with star operators ( )
which are linked to certified specialized multimedia descriptions of the place. Two LML documents are
defined for describing the culture related characteristics for the Italian and the Japanese cultures. Moreover,
two TL documents describe the materialization characteristics of the system for two different platforms: PDA
and laptop.

Figure 4. Valchiavenna portal for Japanese tourists (A) and Italian domain experts (B) using laptops
Figure 4 shows the representation of two possible final instances of the portal, both being part of the same
family of systems. The first one addresses Japanese tourists using a laptop; it: i). respects the cultural
conventions of the Japanese culture (like language, color and shape representations) and ii). provides the end
user with the specific tools needed for a tourist (annotating points on the map by providing the personal
feedback on the visit of the place). A localized instance of the system may be materialized also for Italian
tourists using different types of devices. The second instance addresses Italian domain experts; it: i). respects
the Italian cultural conventions and ii). provides the end user with the specific tools needed by domain
experts (annotating places on the map providing specialized multimedia descriptions of the place). Similarly,
Japanese domain experts may be provided with an instance of the system localized to their culture and
possibly other platforms.
ISBN: 978-972-8939-25-0 2010 IADIS
We argue that nowadays, end users should be supported in the exploration of the problem space they are
facing by web spaces addressing each end users profile. The paper considers as part of end users profile
his/her culture, his/her role in a domain and the platform s/he is using in interacting with the web space. We
describe an abstraction model for designing web interactive systems, seen as web spaces and which can be
localized to the end users profiles, allowing him/her to interact with the system using systems of signs
familiar to him/her. The paper also describes the method to be used in applying the model together with a
concrete example of the model applied.
The authors would like to thank Piero Mussio for the useful and insightful discussions. This work was
supported by the Initial Training Network "Marie Curie Actions.
Alexander, C. 1977. A Pattern Language: Towns, Buildings, Construction. Oxford University Press, Oxford.
Barricelli, B. R., Iacob, C., Zhu, L. 2009. Map-based Wikis as Contextual and Cultural Mediators, Community Practices
and Locative Media workshop at Mobile HCI09. Bonn, Germany. Available online: http://www.uni-
based_wikis_as_contextual_and_cultural_mediators.pdf (Retrieved: August 23rd, 2010)
Barricelli, B. R. et al, 2009. BANCO: A Web Architecture Supporting Unwitting End-user Development, In: IxD&A,
Design for the Future Experience, Vol. 5, No. 6, pp. 23-30.
Borchers, J., 2001. A Pattern Approach to Interaction Design. John Wiley & Sons, Inc, Hoboken, New Jersey.
Bottoni, P. et al, 1997. Defining visual languages for interactive computing, In IEEE Transactions on Systems, Man and
Cybernetics, Part A: Systems and Humans, Vol. 27, No. 6, pp. 773-783.
Brancheau, J. C. et al, 1993. The management of end-user computing: status and directions. In ACM Computing Surveys,
Vol. 25, No. 4, pp. 437-482.
Calvary, G. et al, 2003. A Unifying Reference Framework for Multi-Target User Interfaces, In Interacting with
Computers, Vol. 15, No. 3, pp. 289308.
Carrara, P. et al, 2002. Toward overcoming culture, skill and situation hurdles in human-computer interaction, In
International Journal Universal Access in the Information Society, Vol. 1, No. 4, pp. 288-304.
Carrara, P., Fresta, G., and Rampini, A. 2000. Interfaces for geographic applications on the World Wide Web: an
adaptive computational hypermedia. Proc. of ERCIM Workshop on User Interfaces for all. Florence, Italy, pp. 341-
Costabile, M. F. et al, 2007. Visual Interactive Systems for End-User Development: a Model-based Design Methodology,
In IEEE Transactions on Systems Man and Cybernetics, Vol. 37, No. 6, pp. 1029 1046.
Lin, J. and Landay, J. A. 2008. Employing patterns and layers for early-stage design and prototyping of cross-device user
interfaces. Proc. of CHI 08, Florence, Italy, pp. 1313-1322.
Molina Mass, J. P., Vanderdonckt, J., Simarro, F. M., and Lpez, P. G. 2005. Towards virtualization of user interfaces
based on UsiXML. Proc. of the Conference on 3D Web Technology, Bangor, UK, pp. 169-178.
Treadaway, C. P. 2009. Hand e-craft: an investigation into hand use in digital creative practice. Proc.of C&C09,
Berkeley, USA, pp. 185-194.
Valtolina, S., Mussio, P., Barricelli, B. R., Bordegoni, M., Ferrise, F., and Ambrogio, M. 2009. Distributed knowledge
creation, recording and improvement in collaborative design. Proc. of KES IIMSS 09, Venice, Italy, pp. 31-41.
Wright, P. et al, 1997. Techniques & Tools for Using Color In Computer Interface Design. In Crossroads, Vol. 3, No. 13,
pp. 3-6.
Xu, A., Smith, B., and Bailey, B. 2009. A collaborative interface for managing design alternatives. Proc. of C&C09,
Berkeley, USA, pp. 443-444.
IADIS International Conference WWW/Internet 2010
Rafael Ferreira
, Rinaldo J . Lima
, Ig Ibert Bittencourt
, Dimas Melo Filho
, Olavo Holanda
Evandro Costa
, Fred Freitas
and Ldia Melo

Informatics Center, Federal University of Pernambuco - Av. Professor Lus Freire, 50740-540, Recife Brazil
Web Optimization Group, Federal University of Alagoas - Av. Lourival Melo Mota, 57072-970 Macei, Brazil
College of computers and technology of Pernambuco - Rua do Progresso 441, 50070-020 Recife, Brazil
The development of the Web brought an interactive environment in which an increasing number of users are sharing their
knowledge and opinions. Blogs are a growing part of this environment. Considering the rate at which knowledge is
created daily in the blogosphere, such an amount of information could be used by several applications. However, the
infeasibility of manual information extraction fromblogs demands the development of new computational approaches.
This context is the rationale for the blog crawlers, software programs capable of searching and extracting information
fromblogs. This paper proposes a framework that helps the user in the task of building blog crawlers. The proposed
framework provides further access to several tools, simplifying the overall development of new applications. We also
present an algorithm for locating the textual content within blogs. Finally, we demonstrate the feasibility of our
framework by means of an instantiation of it that achieved a precision and recall of 73.46% and 71.92%, respectively.
Blog Crawler, Framework, Blog, Information Retrieval, Blogosphere.
Over recent years, the evolution of the Web changed the way users interact with information. This evolution
allowed them to become an active part of the Web through new interfaces that permit the creation and
management of textual content in a collaborative way. This phenomenon is known as Web 2.0 [Isotani et al.,
2009]. With the Web 2.0, several collaborative tools were implemented and made available to the public, i.e.,
wikis, forums and blogs.
According to a Technoratis report in 2008 [White and Winn, 2008], there were more than 133 million
blogs. Furthermore, this same report showed that the activity on blogs doubled every two hundred days. Each
blog consists essentially of publications regarding authors thoughts and opinions. The growing number of
interconnected blogs constitutes what is known as the blogosphere. In other words, the blogosphere is the
entire community of blogs on the Web.
Given the amount of knowledge and information being created everyday in the blogosphere, ones could
envisage it as a potential source of knowledge for numerous applications. For instance, it could be used to
analyze users opinions or personal preferences concerning a specific product or trademark. On the other
hand, in order to make a profitable use of this information, it is necessary to properly retrieve and process it
[Chau et al., 2009]. However, with the increasing number of blogs available on the Internet, it is infeasible to
extract all this information manually, thus there is a need of computational approaches for information
extraction from blogs.
Much research has being developed to automate the blog extraction process. There are approaches using
techniques from natural language processing [Berwick et al., 1991], information retrieval [Manning et al.,
2008] and information extraction [Mooney and Bunescu, 2005]. For the document acquisition task, the field
of information retrieval (IR) stands out [Hotho et al., 2005], since it is mainly concerned with identifying
relevant texts for a particular purpose within a huge text collection. However, these IR techniques alone are
not so effective when applied on the blogosphere. A more feasible alternative is to contextualize the posts of
ISBN: 978-972-8939-25-0 2010 IADIS
the blogs. This approach is used by blog indexing services like Technorati, Icerocket and Blogcatalog
companies maintain tagging services for Internet sites and blogs, considerably increasing the semantic level
associated to them [Mathes, 2004]. For instance, if a blog post related to a math class is marked with an
education tag, it is inserted into the educational context, hence facilitating the search process and providing
better search results to the user.
In this light, in order to enable users to perform their searches, the text and information related to each
blog need to be properly indexed and stored. Such a task is performed by blog crawlers [Arasu et al., 2001].
In order to build a blog crawler, we should consider many aspects related to blogs itself, such as language
used by them, blog indexing service, preprocessing tasks, and indexing techniques. In addition, the blog
crawler should be easily extendable, due to the dynamic nature of the Web.
Thus, a good framework for constructing blog crawlers should attend to all these issues [Johnson and
Foote, 1988]. With a framework that could be easily extendable for different applications, the users would
create specific blog crawlers with little effort.
Accordingly, we propose a framework for building context-based blog crawlers. The framework uses tags
to increase the semantic level of the blogs and provides many services, such as preprocessing, indexing and
general text extraction from HTML. We also present an instantiation of the framework based on Technoratis
tagging system.
This article is structured as follows: Section 2 details the frameworks architecture components. An
example of how to instantiate the proposed framework, as well as first results and discussion are shown in
Section 3. Related work is presented in Section 4. Finally, in Section 5, we present our conclusions and future
The proposed system architecture consists of a white-box framework which allows the development of
context-based blog crawlers. This framework provides application services such as text preprocessing, blog
indexing, text extraction from HTML pages, and an API allowing the user to easily implement data
persistence. The general architecture is shown on Figure 1. The aforementioned services are explained in
more detail below.
2.1 The Crawler Module
This is the main module of the framework. The Application and TagParser classes from this module are
directly connected to the frameworks core. In order to create an application with it, the user must extend
these two classes. The user can configure some aspects of the application by initializing a few attributes when
extending the Application class. For example, she can define the language that will be used by the crawler,
the preprocessing algorithm, and the text extraction method, just to mention a few. Concerning the TagParser
class, the only requirement for using it, is to define the tagSearch() method, which executes a search and
returns a list of blogs matching a given tag. This will be further discussed in the Section 3.1.
2.2 The Application and General Services Modules
The Application Services Module provides many services that are used in the crawling process. These
services allow users to create several applications, i.e., in other scenarios, the framework is not limited to the
creation of blog crawlers. It follows a short description of all these services:
Preprocessing: This service removes the information considered irrelevant to the blog analysis. The
current version of the framework can perform four types of preprocessing: i) CleanHTML [Hotho et al.,
2005], which is responsible for cleaning all HTML tags; ii) EnglishFiltering [Frakes and Baeza-Yates, 1992]
which can filter just English text; iii) EnglishStemming [Porter, 1980], performing a classical stemming
technique, reducing the word to its lemma; and iv) WhiteSpace, which removes extra whitespaces;

1, http://www.icerocket.comand, respectively.
IADIS International Conference WWW/Internet 2010
Indexing: We delegated the implementation of this service to the Lucene
tool. The service is in
charge of indexing texts for the search process.
Text Extraction from HTML pages: This service extracts the textual content of blog posts. It is
based on the SummaryStrategy algorithm that will be discussed in Section 2.4;
Persistent storage API: This API is responsible for saving, retrieving and updating instances in
either a MySQL
database or in a RDF schema-based repository as Sesame
The General services module implements basic services which are used in the crawling process. It
provides file manipulation, HTTP requests, XML file manipulation, and language detection. All web
connections are handled by the HTTP service provided by this module.

Figure 1. The framework's architecture
2.3 The toolkit and Persistence Modules
The Toolkit Module has several interfaces to the following set of APIs and tools: i) Lucene and Lingpipe
for extraction and text retrieval; ii) Hibernate and Elmo, which handle data persistence transparently; iii)
HttpClient, for HTTP page retrieval; iv) Google Language Detection, to detect the language of the blogs text.
The Toolkit module makes transparent the use of each one of these tools, providing an effortless access to
them which decreases the learning time. For instance, the user does not to go deeper in details about Lucene
API in order to build an application that actually uses it. Instead, she could delegate to the Lucenes functions
of the framework.
The Persistence Module is responsible for the storage. It supports MySQL databases and Sesame, which
is an open source J ava framework for storage and querying of RDF data. Particularly, if the user chooses to
use Sesame, she has to define the ontology which will be used as the database schema.

Lingpipe is a toolkit for processing text using computational linguistics:
ISBN: 978-972-8939-25-0 2010 IADIS
2.4 The Summarystrategy Algorithm
During the implementation of this framework, we have faced the task of extracting the textual content of
blogs posts. In order to do that, we had to find the delimiters indicating both the beginning and the ending of
the posts within a blog page. Furthermore, the fact that each blog has its own HTML structure makes this
task more difficult. Therefore, we propose the SummaryStrategy algorithm which uses both the link of the
original blog to retrieve it and a brief summary of the post to find its text within a blog page. Since we rely on
blog indexing services like Technoratis to retrieve the summaries of the blogs, their acquisition was a minor
issue. Being the summary text and the text inside the blog exactly the same, the algorithm can make use of
the former as a reference to find the latter inside the blog page. However, the blog summary is seldom written
in plain text (without any formatting markup), contrasting with the text inside the blog and this fact can
prevent the algorithm to locate the summary within a blog page. To overcome that difficulty, if the algorithm
was not able to find the beginning of the posts text content, it tries to use smaller fragments of the summary
as a reference, until it finds it.
The algorithm also has to detect the posts endings. By analyzing the layout tags used by different blog
pages, we noticed that we could use the block-delimiting HTML tags (i.e. </div>) as delimiters of the textual
content of blogs posts. The Figure 2 shows the pseudocode of the SummaryStrategy algorithm, where
Summary and blogHTML are strings, and possibleEndings is a list of strings.

1 while (length of Summary > 0) do:
2 if (Summary is found within blogHTML) then:
3 textStart position of Summary within blogHTML
4 for each ending in possibleEndings do:
5 if (ending is found within blogHTML, after textStart) then:
6 textEnd ending position within blogHTML, after textStart
7 return text between textStart and textEnd
8 end if
9 end for
10 return not found
11 end if
12 delete the last character of Summary
13 end while
14 return not found
Figure 2. Pseudocode of the summarystrategy algorithm.
The algorithm is fast and has low processing requirements. The downside of the algorithm is that it relies
on a summary and on hardcoded delimiters (i.e. the div tag) to extract the full text from blogs posts.
In a future version of our framework, we could implement webpage template-detection techniques such as
proposed by [Wang et al., 2008]. The template-detection algorithms are able to infer the layout template of a
web site, generally by counting block frequencies. Thus, these algorithms can be used to determine the blogs
This section briefly describes the development of an application using the proposed framework, and
discusses an analysis of the developed application.
3.1 Developing an Application with the Framework
As previously stated, the only requirement to create an application using our framework is to extend the
Application and TagParser classes.
When extending the Application class, the user can configure some aspects of the application by setting
attributes of the class. The user can define features as the desired language of blogs, the type of persistence,
IADIS International Conference WWW/Internet 2010
the preprocessing and indexing techniques, and the tags to search for. This gives a lot of flexibility to the
In order to extend the TagParser class, the user only needs to implement the tagSearch() method. This
method is responsible for finding blogs with a specific tag. At first this may seem a difficult task, however it
gets a lot simpler by making use of tagging services as the one provided by Technoratis. Nevertheless, its
worth mentioning that we do not recommend the use of multiple data sources, i.e., using several blog
indexing services at the same time, because this can cause information inconsistency.
In our instantiation we used five categories (tags): education, economy, entertainment, sports and movies.
We have chosen to use a limited number of categories to simplify our analysis. For the tagSearch() method
we used Technoratis search service because it provides a wide range of tags (both general and specific ones).
For instance, there are general tags in politics context and specific tags related to political personalities, as
Barak Obama. In addition, even though the framework allowed us to work with blogs in many languages, for
the sake of simplicity, we preferred just to deal with blogs in English. The SummaryStrategy algorithm was
used for the text extraction from blogs posts.
3.2 Results and Discussion
We evaluated one frameworks instantiation and the results are discussed below. The evaluation was based in
two common metrics of information retrieval systems: precision and recall [Baeza-Yates and Ribeiro-Neto,
1999]. The precision allowed us to know how many of the retrieved posts were relevant. In our context, a
post was considered relevant only if it was complete and written in English. The recall indicates how many
relevant posts were retrieved, considering the total set of relevant posts.
For this first simple analysis, we selected twenty blogs from each of the five aforementioned categories.
The crawling results were inspected manually. Table 1 shows the results.
Table 1. Results by category
Category Precision Recall
Education 61.5% 65%
Average 73.46% 71.92%

As we could notice, some categories performed better in this evaluation than others. For instance, the
sports category got a precision of 87.5% and a recall of 80%, while the entertainment category scored 60%
for both the precision and recall. This variability in the results occurred due to differences in the level of
language in which the blogs are written in. Sports blogs tends to be more formal and better structured,
because they tend to have connections with larger organizations. On the other hand, blogs about
entertainment subjects are mainly written by the general public, they tend to be less formal and contain slang
language that affects the text extraction process.
The algorithm managed to achieve an average precision and recall of 73.46% and 71.92%, respectively.
[Glance et al., 2004] proposed an application for analyzing collections of blog using machine learning and
natural language processing techniques to select the most quoted phrases and accessed blogs.
In [Hurst and Maykov, 2009], the authors proposed a blog crawler that was concerned with specific issues
such as performance, scalability and fast update. Their crawler takes in account low latency, high scalability
and data quality, and appropriate network politeness. The distributed crawler architecture is composed of
scheduling, fetching, and list creation subsystems, as well as a natural language parser.
More recently, [Wei-jiang et al., 2009] tried to increase the quality and accuracy of searches in blogs.
The authors proposed a blog-oriented crawler, and compared it to a topical crawler using their specific
ISBN: 978-972-8939-25-0 2010 IADIS
features. For the development of this blog crawler, it was considered three algorithms for recognizing blog
pages, for blog crawling, and for links extraction.
Despite the fact that all works described above explored important and interesting features of the
Blogosphere, they are not directly concerned with the blog retrieval and blog description issue. For Glance et
al. the blog retrieval is important, however they use metrics that do not increase the description level of the
blogs. Wei-jiang et al. focused on blog searching, considering some aspects of the extraction and information
retrieval issues. Finally, Hurst and Maykov just considered the blogs searching task as the more important
requirement of their system.
In our frameworks instantiation, we used tagging systems to enhance the semantic level of retrieved
blogs, improving the search process. Indeed, all the cited works are focused on one specific domain.
Contrarily, our proposal is more open and general in which the user will have little effort to extend our
framework for creating new applications. In other words, our proposed framework is at a higher level of
abstraction which facilitates the customization of blog crawlers in the context of any greater application.
The main purpose of this work was to propose a framework for building context-based blog crawlers. The
proposed framework comprises a large set of tools and keeps a high level of abstraction for the developer,
providing an easy access to its tools and APIs. Many of its aspects were detailed. The potential of our
framework was evidenced by an instantiation on a typical blog extraction task that achieved satisfactory
results: an average precision of 73.46% and average recall of 71.92%.
The framework presents many expected features suggested by the software engineering field, such as:
reliability, risk reduction, rapid prototype development, and easy maintenance, among others. Furthermore,
our framework provides access to many tools and is able to deal with multiple languages.
As future work, we intend to: (1) perform a deeper evaluation of an improved version of the
SummaryStrategy algorithm (with some more heuristics). In addition, we would like to compare it with other
layout template-detection algorithms; (2) add an information extraction module for enabling further blogs
analysis; and finally, (3) provide a component for combining the tags suggested by several blog indexing
services, based on a fine-grained information extraction performed by the aforementioned module. Moreover,
this same component would be able to classify new untagged retrieved blog pages improving in this way the
robustness of our framework.
Arasu, A. et al., 2001. Searching the web. ACM Transactions on Internet Technology, 1(1), pp. 243.
Berwick, R. C. Abney, S. P. and Tenny, C., 1991. Principle-Based Parsing. Computation and Psycholinguistics. SIGART
Bull, 3 (2), pp. 26-28.
Baeza-Yates, R. A. and Ribeiro-Neto, B., 1999. Modern Information Retrieval.: Addison-Wesley Longman Publishing.
Chau M., LamP. , Shiu B., Xu J. and Cao J . 2009. A Blog Mining Framework. Social Network Applications. IEEE
Computer Society.
Frakes, W. B. and Baeza-Yates, R. A., 1992. Information Retrieval: Data Structures & Algorithms. Upper Saddle River:
Glance, N.S. Hurst, M. and Tomokiyo, T., 2004. Blogpulse: Automated trend discovery for weblogs. In: WWW2004,
workshop on the weblogging ecosystem: aggregation, analysis and dynamics. New York City, USA, 17-22.
Hotho, A. et al., 2005. A Brief Survey of Text Mining. GLDV Journal for Computational Linguistics and Language
Technology, 20 (1), pp.19-62.
Hurst, M. and Maykov, A., 2009. Social streams blog crawler. In: Proceedings of the 2009 IEEE International
Conference on Data Engineering, Shanghai, China.
Isotani, S. et al., 2009. Estado da arte emweb semntica e web 2.0: Potencialidades e tendncias da nova gerao de
ambientes de ensino na internet. Revista Brasileira de Informtica na Educao, 17 (1), pp.30-42.
IADIS International Conference WWW/Internet 2010
J ohnson, R. E. and Foote, B., 1988. Designing reusable classes. Journal of Object-Oriented Programming, 1 (2), pp.22
Manning, C. D. et al., 2008. Introduction to Information Retrieval. Cambridge: Cambridge University Press.
Mathes, A., 2004. Folksonomies-cooperative classification and communication through shared metadata. Computer
Mediated Communication.
Mooney, R. J. and Bunescu, R. C. 2005. Mining knowledge from text using information extraction. SIGKDD
Explorations, 7 (1), pp.3-10.
Porter, M. F., 1980. An algorithmfor suffix stripping. Program, 14 (3), pp.130-137.
Wang, Y. Fang, B. Xeng, C. Guo, L. Xu, H., 2008. Incremental Web Page Template Detection. In: WWW2008: 17

International World Wide Web Conference. Beijing, China 21-25.
Wei-jiang, L., Hua-suo R., Kun H., J ia L., 2009. A New Algorithm of Blog-oriented Crawler. In: International Forum on
Computer Science-Technology and Applications, IFCSTA.
White, D. and Winn, P., 2008. State of the Blogosphere [online] Available at:
<>[Accessed 29 August 2010].
ISBN: 978-972-8939-25-0 2010 IADIS
Juliana de Melo Bezerra and Celso Massaki Hirata
Computer Science Department, ITA
S.J.Campos, Brazil
In virtual communities, norms are established in order to discipline and organize their members. The community is called
self-organizing virtual community when members are responsible to define and update the norms. Cooperation is
required both in the execution of the community activities and in the definition of norms. As consequence of cooperation,
conflicts may arise, and as consequence of conflicts, norm infractions may occur. Norm infractions can be also
consequence of vandalism and misunderstanding of norms. Dealing with norm infractions is essential to keep the order in
the community. An effective way to deal with norm infractions is provide approaches of enforcement of norms. This
article presents an extension of self-organizing virtual community model and a process for enforcing the norms. The
model considers malicious members, enforcement norms, and enforcement procedures. The process takes into account
norm infractions detection and analysis, sanctions, and damages. The proposed model can be used as a guideline for
designing and constructing self-organizing virtual communities.
Self-organizing, virtual community, enforcement, norm, sanction, damage.
A community is an association of people, called members who are oriented by individual motivations and
cooperate in order to achieve a common objective (Lenhard, 1988). When the relationship among members
and the cooperation itself takes place over the Internet, the community is known as virtual community
(Rheingold, 1998). The term virtual is applied because the members are not in contact personally, but they are
connected through the Internet.
Norms are established in order to discipline and organize the members in communities. A norm is a type of
principle or rule that states obligation, permission, power attribution or competence attribution (Palaia, 2005)
(Lyons, 1984). As members interests and demands change over time, virtual communities are not fixed and
consequently norms evolve. Virtual communities that promote their own evolution are called self-organizing
virtual communities because they create, increment and adapt the norms that govern their relations and
processes. The self-organization stimulates the participation and involvement of members, contributing to the
persistence of the community over time (Moe et al., 2008).
Examples of self-organizing virtual communities are (Lattemann, 2005), Wikipedia
(Goldspink et al., 2008) (Beschastnikh et al., 2008), (Jensen and Scacchi, 2005) and
(Lattemann, 2005). In these communities members can change the norms by participating on their
enhancements. According to the above definition, virtual communities as Linkedin and Orkut are not self-
organizing ones, because norms are established by a board (that maintain the community) and not by
The norms definition is essential for the community organization and the enforcement of such norms shall
be also addressed. As explained by Palaia (2005), a norm shall contain a coercion element in order to force the
members to follow it: if a member disregards a norm, he/she suffers a sanction effect, which is a punishment
for the violation. It is important to mention that a sanction is not always a punishment. The punishment is a
negative sanction. There can be also positive sanctions, as rewards (Palaia, 2005). The positive sanctions are
useful to instigate the effort of members in some direction, for example a reputation scheme to congratulate
IADIS International Conference WWW/Internet 2010
members that make relevant contributions to the community. In order to ease the understanding of the text, we
use the term sanction as a synonym of punishment.
The enforcement of norms in self-organizing virtual communities has attracted interest. Nahon and
Neumann (2005) study how virtual communities that share messages, specifically forums, regulate themselves.
They explain that some members called managers are responsible to detect and disregard messages that
infringe norms. Geiger and Ribes (2010) relate a case of banning a vandal in Wikipedia. They explain how
some members, the vandal fighters, work assisted by tools called bots (Geiger, 2009) in order to keep the
community harmonious by enforcing the norms. The article gives an overview about how Wikipedia enforces
its norms by combining people and bots. Forte and Bruckman (2008) identify in Wikipedia the design
principles for self-organizing communities studied by Ostrom (2000) for physical communities that manage
natural resources. The principles include concepts of monitoring, sanction, and enforcement. Other articles
Flynn (2001), Moe et al. (2008), and Lattemann and Stieglitz (2005) report on how the enforcement of norms
is used in some specific communities. However the articles do not provide a general model for the
enforcement of norms in self-organizing virtual communities.
The enforcement of norms helps to assure that the norms are obeyed by members, contributing to keep the
order in the community. Keeping the order, members cooperate harmoniously and they are more confident on
the community ability to handle undesired situations. Therefore, community credibility and members
motivation increase. In this article, we propose an extended model of self-organizing virtual communities
addressing the enforcement of norms with specific norms and procedures. In order to accomplish the
enforcement, we also propose an enforcement process taking into account norm infractions, sanctions and
In Section 2 we discuss some aspects that lead to the existence of norm infractions and we address the
concepts related to a norm infraction. In Section 3, we propose the extended model to deal with enforcement in
virtual communities. We identify the enforcement process and then we include it in the model of self-
organizing virtual communities. Section 4 provides some discussions with respect to the model. The last
section concludes our work.
In this section, we discuss the cooperation and competition forces in communities and norm infractions.
Besides we describe some concepts associated to a norm infraction.
2.1 Norm Infractions
In all communities, physical or virtual, there are forces that join and dissipate members. A force is called
cooperation, when members work together aiming a common objective. In virtual communities cooperation is
aided by tools of computer supported cooperative work. In order to organize the cooperation, members are
assigned to roles that define permissions on activities that they are allowed to perform (Lenhard, 1988). So the
operational norms are established in the community.
Cooperation is a motivation for the existence of a community; however it is not the unique force in the
community. There is the competition motivated by opposite views between members. A competition is not just
for obtaining a scarce good or the best of the goods, but it may happen to conquer power, fame, status and
partners (Ogburg and Nimkoff, 1972).
Competition sometimes contributes to incentive members in their activities. However, when competition
amplifies and the objective of a member becomes to win the other, a conflict arises (Lenhard, 1988). Norm
infractions can appear during a conflict, because the opponents are involved in the dispute and may ignore the
established norms that govern behaviors and relations. In this case, norm infractions are related to insult and
personal attacks. In some virtual communities the possibility of occurrence of conflicts is reduced by the lack
of cooperation, for example in communities whose objective is the social networking, such as Orkut and
Linkedin. In this kind of community, a member maintains his goods and shares them only with desired
members. Members do not develop an activity together, so cooperation is almost inexistent as well as conflict.
Other virtual communities can reduce conflicts even with the presence of active cooperation. It is the case
of JEMS (Journal and Event Management System), a community of researchers that submit and review papers
ISBN: 978-972-8939-25-0 2010 IADIS
for technical conferences. In JEMS the profiles of some members (reviewers) are kept in anonymity during
discussions, which may contribute to the reduction of personal attacks and offenses.
Self-organizing virtual communities are more prone to conflicts, because the cooperation is required both
in the accomplishments of the community activities and in the definition of norms. In both cases, norm
infraction may occur as consequence of conflicts, so the community has to resort to the enforcement of norms
so that the order can be maintained.
Norm infractions may arise during conflicts; nonetheless conflicts are not something that should be
avoided in communities. Conflicts are useful because they generate distinct points of view, which can help
members to better analyze the topics being discussed. It should be noted that norms infractions are not only
consequence of conflicts, they can be caused by either vandalism (when a member intentionally desires to
corrupt community resources) or some misunderstanding during the interpretation of norms.
2.2 Concepts related to Norm Infraction
Sanctions guide the behavior of the members so that norms established by the community are obeyed. Clear
and unambiguous norms must be provided in order to reduce misunderstandings. We identify three concepts
related to norm infraction: sanction, damage and impact. Figure 1 shows the concepts and some of the
A norm infraction causes damages in the community, for example misuse of a page editor makes available
contents with offenses. Damages have impact, such as the undesirable exposition and loss of members, and the
decrease of community credibility. A sanction, applied to a transgressor, should depend on the impact of the
caused damage, for example a sanction for writing nonsense phrases in content should be less severe than a
sanction for writing offenses. A sanction may also depend on the history of the transgressor, for instance a
member that insists on vandalism has to be punished in a harder way in order to change his/her behavior.

Figure 1. Concepts related to a norm infraction
The damages have to be recovered in order to minimize their impacts. In virtual communities based on
forums, the damage is recovered by disregarding messages. Examples of damages in forums are: irrelevant
messages (the messages are not related to the community objective), messages that infringe community culture
(offenses to members) or that contain commercial information (spam) (Nahon and Neumann, 2005).
Priedhorsky et al. (2007) study types of damages in Wikipedia and their impact. Some reported damages
are misinformation (or false information), mass and partial delete (to delete articles content), offenses (to
write text with obscenities, hate speech, and personal attacks), spam, and nonsense (text that is meaningless to
the readers). Some impacts of the damages are to confuse, offend or mislead readers. Besides, the impacts may
include the loss of members and the reduction of articles quality. A common technique to recover the
damages is the reversion of the contents to a valid version.
Lattemann and Stieglitz (2005) explain that in open source communities the sanctions are not limited to the
exclusion of the transgressor and the loss of reputation. Transgressors may be flamed, meaning they are
publicly named and judged by other community members. In this kind of community, the damages are mainly
concerned about misconduct in the developed codes. The authors also comment that the threat of the sanctions
may help to achieve quality assurance.
Flynn (2001) relates the case of an adult-oriented forum that had norms stating that it was forbidden to
share child pornography and to practice spam. The community only recovered the damages caused by
transgressors by removing the inadequate messages. However no sanction was applied to the transgressors, so
there were no means to block them, because the members did not use login and password to access the system.
As the number of vandals increased and the community did not find a way to stop them, the forum was closed.
This case illustrates the need of sanctions to enforce norms in virtual communities.
IADIS International Conference WWW/Internet 2010
In this section, we extend the model of self-organizing virtual communities with the enforcement of norms.
We also propose an enforcement process, detailing its activities and relations.
3.1 Enforcement in the Community Model
Bezerra and Hirata (2010) propose a model for self-organizing virtual communities. The model is composed
by three main elements: Members, Norms and System. The Members belong to the community motivated by
their personal goals. They define and improve Norms that possibly change how they work by imposing
discipline. The System, by implementing the norms, provides the means for the members to work, interact, and
achieve their goals. The System comprises Resources, Access Control policy model and Procedures.
Resources encompasses all the goods developed in the community, such as messages in forums, articles in
Wikipedia, and documents and codes in open source communities. The permissions over the Resources are
handled by an Access Control policy model. Procedures are driven by the norms and detail the operational
participation of the members, for example a procedure to invite a member for the community, and a procedure
to revise an article or code.
In this article we extend the previous model with elements related to the enforcements of norms as shown
in Figure 2. The norms and procedures related to the cooperation of members to develop their Resources are
now called, respectively, Work Norms and Work Procedures. There can be distinct kind of Work Norms, for
example in Wikipedia they are mainly divided in content-related and behavior-related norms (Geiger and
Ribes, 2010). The way that the enforcement is performed has to be also established using additional norms we
call Enforcement Norms. The Enforcement Norms regulate how the community detects violations in norms
and deals with the concepts of sanctions and damages. Besides, the Enforcement Norms have associated
procedures called Enforcement Procedures.

Figure 2. Self-organizing virtual community model with enforcement of norms
The extended model also includes a new relation called references between system and norms, because
during the enforcement activities, the Enforcement Norms may be referenced to. Another consideration is that
the Access Control can be seen as the primary mean of enforcement, because it assigns permissions to
members manipulate resources according to the rights and duties established by norms.
The model shown in Figure 2 divides the members into two classes: non-malicious members and malicious
members. In the former model proposed by Bezerra and Hirata (2010), all the members are considered non-
malicious. The non-malicious members use the system correctly obeying the defined norms. The malicious
members misuse the system by infracting the established norms. As a consequence of the malicious behavior,
the mechanisms to enforce norms may make use of the system to restrict the transgressor activities.
In order to know what the Enforcement Norms and Procedures should address, the next section identifies
some activities related to the enforcement of norms. In this article, the term activity is used to indicate an
ISBN: 978-972-8939-25-0 2010 IADIS
abstract task or set of tasks. The term procedure is a concrete and detailed way to accomplish the activity, i.e.
its task or set of tasks, considering the specificities of the community established in norms.
3.2 A Process for the Enforcement of Norms
In Figure 3 we propose a process for the enforcement of norms in self-organizing virtual communities. The
process is quite general and it is based on the Incident Response Life Cycle proposed by NIST (2008) to
handle computer security incidents. Besides, the process considers the concepts of sanction and damage. The
process identifies the following activities: Monitoring, Analyses, Sanction Application, Damage Recovery, and
Unexpected Event Handling.
In order to explain the activities we make use of a case study, reported by Geiger and Ribes (2010) that
discusses the banning of a vandal in Wikipedia. In the case study, a user committed several norm infractions
by vandalizing articles (music albums), such as Before we Self Destruct, The Mirror and 808 &
Heartbreak. The editors that are responsible for the enforcement of norms in Wikipedia are called vandal

Figure 3. Process for the enforcement of norms in self-organizing virtual communities
The Monitoring activity allows identifying events over the Resources in order to detect norms infractions.
It is related to some Work Norms. For example, assuming a Work Norm that forbids content with attacks and
offenses to members, one can define a list of insulting words to be searched during the Monitoring.
In Wikipedia, the Resources are articles, while the monitored events are the suspicious editions made in
articles. In the case study, the transgressor made an offending edition (for instance, using improper words) to
the album Before we Self Destruct. The edition triggered various vandalism-detection algorithms, for
example Huggle. Huggle is tool or bot that monitors automatically articles in Wikipedia. Other tools include
the popular Twinkle, and the less known, ClueBot and Lupin (that search obscene words and commonly
misspelled words). Although in the case study the monitoring was made automatically, members may be
involved in the identification and reporting of a vandal.
The purpose of Analyses activity is to examine the incident considering the community norms in order to
classify the incident as a norm infraction or not. In Wikipedia Huggle also helps in this phase, because the
provided rank of incidents already includes some decision criteria such as: the kind of the user who made the
edit, if the user had already an edit reverted, and if the user has any warning about norm infraction.
If the incident is a norm infraction, the Enforcement Norms may be used to determine the sanctions to be
applied to the transgressor, as well the recovery of the damages. In the case study, the incident was clearly a
norm infraction, because of the use of obscenity words. The damage was undone by an editor who reverted the
article using Huggle tool. No sanction was applied.
Later the same transgressor vandalized The Mirror album which generated an incident. At that time, an
editor using Twinkle tool both reverted the article and sent a warning message to the transgressor. The warning
mechanism makes use of user talk pages, which are public wiki pages created by the system for each user. The
user talk pages have become a kind of database of norm infractions for particular users, which is useful during
the Analysis phase.
Regarding the warnings in Wikipedia, there are some templates specifically to inform the transgressor that
the edit in question was not encyclopedic. They are divided into four levels of severity. For example, in the
case study the first warning was One of your recent edits, such as the one you made to The Mirror (album)
did not appear to be constructive and has been reverted. The fourth-level warning sent to the same
IADIS International Conference WWW/Internet 2010
transgressor was: You will be blocked from editing the next time you vandalize a page, as you did with this
edit to 808s & Heartbreak.
In Wikipedia, the hardest sanction is the block of the transgressor. This sanction is only applied by the
administrators who have the technical and social authority to issue temporary blocks. However the editors that
analyze the incidents have the possibility to make formal ban requests to administrators using the AIV
(Wikipedia: Administrator intervention against vandalism) page. In the case study, the transgressor was
indicated to AIV and, after one more vandalism, he was blocked by an administrator.
Tools, as Huggle and Twinkle, aid the vandal fighters during the enforcement process in Wikipedia. The
tools both monitor events and help to analyze the incidents considering the profile and history of transgressors.
They also support the damage recovery and may suggest sanctions to be applied to the transgressor based on
predetermined list; however the decision is always made by the editors who operate the tools.
Not always an incident characterizes a norm infraction. An incident can be generated due to an error or
wrong configuration of some automatic components, for example components that support the Monitoring
activity. An incident can occur because a norm is ambiguous and a member did not understand it.
Furthermore, an incident can occur because the subject is not addressed by any norm. In these cases, an
incident is called an unexpected event and has to be managed. The Unexpected Event Handling activity deals
with unexpected events, for example, the procedure may include the correction of the faulty component and
the proposition or update of norms.
In order to perform the enforcement activities, some roles in the Access Control of the community system
can be identified. The Monitoring can be performed by resource administrators that are responsible to
guarantee the integrity of the community resources. A committee of resource administrators can conduct the
Analyses activity. The Sanction Application activity can be executed by user administrators that configure the
permissions of the users over the resources. The Damage Recovery activity can be performed by resource
administrators that recover the resource to a valid version. Depending on the damage severity, such as denial
of service, the support of technical analysts can be needed to recover the corruption. Due to the unforeseen
nature of the Unexpected Event Handling activity, all the mentioned roles can participate: a resource
administrator can update a resource configuration, a user administrator can correct a wrong member profile,
and a technical analyst can improve an automatic component. Besides, in the Unexpected Event Handling
activity, norm administrators can be necessary to coordinate the suggestions of new norms and the corrections
of established norms, by guaranteeing the community consensus.
With respect to access control, for example, Wikipedia defines some roles related to enforcement:
administrators, bureaucrats and stewards. There are specific permissions regarding sanction application, such
as block (to block an IP address, user name or range of IPS from editing) and userrights (to change the user
groups of a user). There are also permissions related to damage recovery, such as purge (to purge a page) and
rollback (to revert a bad editing).
The enforcement activities are regulated by the Enforcement Norms and are implemented as detailed
operations using the Enforcement Procedures. The procedures are performed by members of the community
assigned to specific roles in the Access Control, in general known as managers or administrators. The
Enforcement Procedures can make use of tools to assist the managers. So, managers and tools work together
to accomplish the whole enforcement process. In general managers are volunteers as all members in self-
organizing virtual communities. These managers use the system to perform the enforcement operations;
therefore, they must be seen as non-malicious members in Figure 2. The process to become a manager can be
very rigorous, as occurs in Wikipedia (Forte and Bruckman, 2008).
The Monitoring can be made manually by members of the community or it can be made automatically
using bots, as Wikipedia does in some articles editions (Geiger, 2009). The choice between humans or bots
depends basically on the possibility of an automatic detection and its development cost. Priedhorsky et al.
(2007) comment about the difficult to detect automatically some infractions in Wikipedia, for instance the
detection of misinformation (article with false information) is almost impossible, because it requires
understanding the article content.
ISBN: 978-972-8939-25-0 2010 IADIS
In both cases, manual or automatic Monitoring, an important concern is to not discriminate any kind of
members, avoiding the predisposition to suspect of the newcomers in community, for example by only
monitoring the events made by the newcomers and not the ones made by the administrators. If a manager
abuses of the power conferred to him/her, it configures a norm infraction and has to be submitted to the
enforcement process too. In this case, the manager must be seen a malicious member in Figure 2.
In some communities the Monitoring can be performed in a subset of events, and not in all events, for
example in the managers relate the impossibility to review all messages. In order to help
the Monitoring, self-organizing virtual communities may specify a norm that confers all members the right to
report violations and abuses, so all the members become guards of the community.
In open source communities, the code development process has embedded a test phase that is a kind of
Monitoring. During the test phase, the detected malfunctions in general are not consequence of norm
infractions. It may occur because only authorized members have access to codes and the process to receive
such an authorization can be much elaborated. For example, uses meritocracy, which means that
the member has first to be an active contributor, so he/she can be indicated by other member for assuming a
role and finally submitted to community voting. Comparing to Wikipedia, where every member can alter the
articles, the possibility of article corruptions can be higher than code corruptions in open source communities.
Due to its nature of decision making, the Analysis activity is mainly made by authorized members and not
only by bots. If it is required by the Enforcement Norms, the Analysis activity may also take into account the
history of the incidents. In Wikipedia, for example, the history of a member is registered in his/her public user
talk page (Geiger and Ribes, 2010). uses the called books and records to keep all the accounts and
proceedings of the members, so any member has the right to examine its information.
In virtual communities the kind of sanction may range from a simple warning to the expulsion of the
community. The warning is seen as an educative sanction, considering that transgressor is a member learning
how to work in the community. However, if the transgressor continues to violate the norms repeatedly, harder
sanctions can be applied as the access denial to resources, which is implemented using the Access Control of
the community system.
The damages in the community resources can be in majority recovered: messages are removed from
forums, articles are reverted in Wikipedia and codes are fixed in open source communities. However damages
caused to other members, for example offenses, are more difficult to deal with and can require the definition of
compensations, such as apologies.
The proposed enforcement process shows the main enforcement activities in an ordered flux; however
some feedbacks can occur in the activities execution. For instance, during the execution of the Analysis
activity, some specific information may be needed, so additional data may be required to be collected in the
Monitoring activity.
In general, the existing self-organizing virtual communities include some sort of enforcement procedures.
However, the procedures are not organized and there is no model to place them, so it is not clear how to
consider the procedures and the relations in the context of members, norms, procedures, access control and
As competition is an active force in communities, conflicts may arise between members or between a
member and the whole community. In a self-organizing virtual community conflicts are more likely, because
members do not only cooperate to accomplish their work activities, but they cooperate to define and maintain
the norms that regulate the community. Conflicts may result in some norm infractions, when the involved
members do not obey the established norms that govern behavior and relations in community. Norm
infractions can be also consequence of vandalism. Therefore the enforcement of norms is of relevance to be
In self-organizing virtual communities, norms have first to be well defined. The community shall also
define norms specifically to govern the enforcement, the Enforcement Norms. The Enforcement Norms are
implemented in the system using Access Control and Enforcement Procedures that detail the operational
tasks. So the procedures are performed by members assigned to administrative roles and possibly assisted by
tools in the system.
IADIS International Conference WWW/Internet 2010
In this article, we propose an extended model in order to guide self-organizing virtual communities to
address the enforcement of norms. We also propose a process with enforcement activities. With the activities
of the enforcement process, Monitoring, Analyses, Unexpected Event Handling, Sanction Application and
Damage Recovery, the community is able to detect an incident and classify it as a norm infraction. In case of a
norm infraction, a sanction is applied to the transgressor and the damage can be recovered. The occurrence of
an unexpected event is also dealt, which allows the enhancement of the enforcement process itself, by updating
tools, procedures, and norms.
The proposed model can be used to analyze the enforcement performed by existing virtual communities, in
order to guarantee that norms are being followed and possibly suggest new directions to it. The proposed
model can also be used to address enforcement of new communities, because its process outlines the basic
activities that help the definition of specific norms and procedures to achieve the desired enforcement.
As future work, we intend to investigate how conflicts can be managed in self-organizing virtual
communities, i.e. how they emerge and can be detected, and how to address them.
REFERENCES Bylaws of The Apache Software Foundation.
Bezerra, J. M. and Hirata, C. M. 2010. Self-Organizing Virtual Communities Model. Submitted.
Beschastnikh, I., Kriplean T. and McDonald, D. W. , 2008. Wikipedian Self-Governance in Action: Motivating the
Policy Lens. Proceedings of International Conference on Weblogs and Social Media.
Flynn, T. R., 2001. Ethics, Law and Technology: A Case Study in Computer-Mediated Communication. Proceedings of
International Symposium on Technology and Society.
Forte A. and Bruckman, A., 2008. Scaling Consensus: Increasing Decentralization in Wikipedia Governance.
Proceedings of Hawaii International Conference on System Sciences. IEEE.
Geiger R. S. and Ribes, D., 2010. The Work of Sustaning Order in Wikipedia: The Banning of a Vandal. Proceedings of
Computer Supported Cooperative Work (CSCW). ACM.
Geiger, R.S., 2009. The Social Roles of Bots and Assisted Editing Programs. Proceedings of International Symposium on
Wikis and Open Collaboration (WikiSym). ACM.
Goldspink, C., Edmonds, B. and Gilbert, N., 2008. Normative Behaviour in Wikipedia. Proceedings of 4
Conference on e-Social Science.
JEMS. Journal and Event Management System.
Jensen, C. and Scacchi, W., 2005. Collaboration. Leadership. Control, and Conflict Negotiation and the
Open Source Software Development Community. Proceedings of Hawaii International Conference on System
Sciences. IEEE.
Lattemann S. and Stieglitz, S., 2005. Framework for Governance in Open Source Communities. Proceedings of Hawaii
International Conference on System Sciences. IEEE.
Lenhard, R., 1988. Sociologia Geral. Livraria Pioneira Editora, So Paulo. LQ Rules.
Lyons, D., 1984. Ethics and the rule of law. Cambridge University Press.
Moe, N. B., Dingsyr, T. and Dyb T. 2008. Understanding Self-organizing Teams in Agile Software Development.
Proceedings of the 19th Australian Conference on Software Engineering. IEEE, pp 76-85.
Nahon K. B. and Neumann, S. , 2005. Bounded in Cyberspace: As Empirical Model of Self-Regulation in Virtual
Communities. Proceedings of Hawaii International Conference on System Sciences. IEEE.
NIST (National Institute of Standards and Technology), 2008. Scarfone, K., Grance, T. and Masone, K. Computer
Security Incident Handling Guide. Publication 800-61.
Ogburg, F. and Nimkoff, M. K., 1972. Cooperao, competio e conflito, in Homem e Sociedade, F. H. Cardoso and O.
Ostrom, E., 2000. Collective action and the evolution of the social norms. In Journal of Economic Perspectives, Vol 14,
No 3, pp 137-158.
Palaia, N., 2005. Essential Notions of Law. Saraiva Publisher.
Priedhorsky R. et al., 2007. Creating, Destroying, and Restoring Value in Wikipedia. Proceedings of Conference on
Supporting Group Work (GROUP). ACM.
Rheingold, H., 1998. Virtual Communities, in The Community of the Future, Jossey-Bass, San Francisco.
Wikipedia Rules.
ISBN: 978-972-8939-25-0 2010 IADIS
Diogo Arantes Fonseca Gonalves da Cunha
, Prof. Artur Ferreira da Silva

and Prof. Jos Figueiredo
*Instituto Superior Tcnico Taguspark - Av. Prof. Dr. Anbal Cavaco Silva 2744-016 Porto Salvo, Portugal
**Instituto Superior Tcnico - CEG-IST
In this paper we explore a strategy to implement an organizational wiki, using a Wikipedia model. We address models to
cope with organizational change and organizational learning within a knowledge management strategy. The results were
positive as our wiki model is now integrated in the day-to-day routines of the company.
Wiki, Wikipedia, knowledge management, organizational change, organizational learning.
As the world evolves into a crescent connected space, where the virtual context is integrated with the
physical reality, organizations need to develop skills to take advantage of this evolution.
On the organizational reality the issue of knowledge management is becoming more important, as
managers try to retain the knowledge of the workers and these workers feel the need to continually improve
their skills (Drucker, 1999). On the public social reality, knowledge sharing and knowledge creation evolved
tremendously in the last years and models have been created that can, and should, be used in other contexts.
The Web 2.0 revolution brought new ways of communicating and working collaboratively, and the results of
this revolution should be harvested to improve organizations.
One of the most popular examples of a public model of knowledge sharing is Wikipedia. But its success
on the public domain doesnt necessarily imply that it would succeed on the much smaller and controlled
organizational environment; probably some adaptations would need to be made to the model and to the
Having that in mind, we set out to investigate if an organization with highly literate workers in IT
technologies would be capable of assimilating a wiki model. More than that, would an internal Wikipedia
available only to the workers, add value to the organization? Although this seems a simple question, the
implementation of such a task entails a considerable amount of unpredictability because the overall process is
heavily dependent on human behaviour. A wiki is a voluntary system of content creation and, even though
Wikipedia quality standards apply to the organizational context, an organization is a complex set of many
different wants and participative patterns that cannot be assured by regulations, and maintaining Wikipedia
quality needs investment in time and resources from the organization that wants to adopt it.
According to Morgan (2006), our perception of an organization is a key factor in its functioning processes.
He says we use metaphors to help our understanding and to better manage our companies, and that
representation is a fundamental topic for an organization to thrive in the market.
Lets consider the metaphor of the living organism applied to the organization entity. A living organism
needs to adapt to its environment in order to survive and is composed of subsystems that interact together to
sustain the survival of the larger system. True evolution is only possible with a correct alignment of all the
subsystems and the environment.
IADIS International Conference WWW/Internet 2010
Following this perspective, we present the Learning Organization integrated architecture of systems
thinking as explained by Senge (Senge, 1990). This concept doesnt define the company that is just able to
learn, it defines the company that has learning processes as the foundation of its development, investing in
areas like innovation, communication and colaboration. Nowadays, the organizational context changes very
quickly and agressively, and companies need to invest in these capabilities so they can keep up with market
evolution and the competition.
After describing the five disciplines that need to be taken in consideration for an organization to become a
Learning Organization (Mental Models, Personal Mastery, Team Learning, Shared Vision and Systems
Thinking), Peter Senge also details which kind of learning disabilities may be causing difficulties in the
change process. These disabilities should be investigated carefully and attentiously, as they may be the
reason why a transformation process takes more time than it should, or doesnt happen at all. Usually these
disabilities are associated with cultural characteristics of the organization, the department or group, as they
are referenced to human behaviour. But, althought identifying some of these disabilites may help, there are
other disabilities that may not be listed and explained.
Some time later, Peter Senge and other coleagues took a different approach to the same problem in
(Senge et al, 1999). They kept standing with the systems thinking model but, instead of focusing on the
intrinsic disabilities that could prevent an organization from changing they focused on the reinforcing
processes that need to be adressed in order to sustain the emergence of change. These processes seem
obvious after understanting their nature and are bound to the ideological perspective of the change, the
deepth of structural innovation needed and the methods and theories used to achieve the intended change.
There are three reinforcing processes to a change model: the process of personal results, that refers to the
passion that a person puts into his job when he loves what hes doing resulting in a positive impact in his life;
the process of networking of commited people, which is aligned with the concept of CoPs (Communities of
Practice) as it refers to the ability that any individual has to transcend himself when genuinely integrated in a
colective; and the process of businness results, that is simply the fact that, good businness results boost up
morale and improve ongoing businness processes.
Nonetheless, althought these reinforcing processes have the objective of sustaining change, they arent
very simple to sustain themselves. According to the author (Senge et al, 1999), there are many challenges that
probably will appear when a change of this nature occurs, and he divides them into three type of challenges:
initiating challenges, sustaining challenges and re-designing challenges. For example, the first challenge of
initiating change is called Not Enough Time and it adresses the problem of the lack of time that most of the
workers have to create, participate or to develop change initiatives. In this project we found some of these
2.1 Leadership
Leadership is a concept that in a colective perspective only started being used recently. It is not a single
characteristic of a person that makes him able to lead others. Certain characteristics, like confidence,
adaptability, and practic inteligence seem to be valuable for this function. But, and more important,
leadership is now seen as the fruit of an entaglement between context and the characteristics of a person. A
person in a particular situation and for a certain group can be the best leader, in a different situation or group
may be not good at all (Mann, 1959).
As such, we need to change our perspective. We need put aside the mith of the hero CEO that saves the
company and brings peace to the workplace (Senge et al, 1999) and embrace a more distributed and emergent
approach to the issue, where communities are made of several leaders, each one with its own situational
ability in the group.
Leadership drives organizational change processes and, organizational change only occurs when there is
leadership supporting and sustaining the iniciatives. Thus, leaders are indispensable for a solid transformation
within an organization.
A functional evaluation of types of leaders is described in (Senge et al, 1999) three types of functional
leaders can be found on the organizational context:
Local line leader, refers to the leader that has the power, position or informal leadership to put
change in action in its practical sense and, that has the ability to assess its results on the making.
ISBN: 978-972-8939-25-0 2010 IADIS
Network leader, also known as internal networker or community builder, is the leader that through
his mobility within informal networks can help create new relationships, aliances or ideas, and help
disseminate new mentalities and attitudes.
Executive leader, is the leader in charge of the administrative and strategic actions within an
organization, assuming the bottom-line responsability for its performance and development that can support
and leverage the change process when acting consistently as "champion" of it.
There are three types of leader behaviours described by (Lewin, Lippitt, & White, 1939):
Authoritarian behaviour, when all decisions are centered on the leader in charge with little or no
regard to group suggestions.
Democratic behavior, when decisions are made considering the thoughts and suggestions of the
Laissez-Faire behavior, which relates to the leader that doesnt impose no control or protocol,
leaving group members to individually decide upon their actions.
We cannot say that a type of behavior is better than the other, but we can find a more suited beahavior to
a specific environment or goal.
The concept of Enterprise 2.0 relates to a use of Web 2.0 systems, such as social networks, wikis, blogs or
other collaborative tools or platforms, within the organizational environment, focusing on the integration of
technical and social facets of the organizational life.
A company that doesnt invest on Enterprise 2.0 systems risks becoming outdated and less competitive
(McAfee, 2006). To support this view we need to clarify that Web 2.0 tools in order to work correctly must
be considered as sociotechnical systems and not just as "tools". There are many types of 2.0 systems that may
be used and they serve different purposes, so there is a need to evaluate the system according to its
functionalities and assess if it is aligned with the company objectives.
The SLATES mnemonic was coined by Prof. McAfee to help on acquisition or development of these
types of systems. Hinchcliffe (2007) wrote a blog post where he added a few characteristics to SLATES,
transforming the mnemonic into FLATNESSES:
Free-form the system must have a dynamic way of organizing content;
Links content has to be linkable;
Authoring user actions must be traceable;
Tags content should be categorized using folksonomy;
Network-Oriented the system must be oriented to a network environment;
Extensions new information about the user needs should be achievable by crossing data;
Search content must be searchable;
Signals the system must provide a notification system, such as RSS or ATOM;
Emergence the system architecture must make content emerge depending on users needs;
Social the system must provide a way to create social relations between users.
Although there are these specifications on Enterprise 2.0 system functionalities, implementing such a
system is not simple. These systems have a voluntary nature, so users are not forced to adopt them. This way,
implementation should be accompanied by a cultural adaptation to the new ways communicating.
Communication, collaboration and integration are the keywords in Enterprise 2.0 initiatives, as workers
have to develop these skills in order to fully align themselves with the philosophy of Enterprise 2.0 and
generate organizational value.
3.1 Wikis
The wiki is a Web 2.0 system that suites the needs of collective authoring of content in a simple, straight
forward process of edition and creation of html pages. The most common functions associated with wikis are:
editing and creating html pages; internal linking between content; future linking to pages that havent been
created; markup language to easily format html pages; version control system; and permission system.
IADIS International Conference WWW/Internet 2010
Although these functionalities may be common to most of the wiki platforms, there are other functionalities
that are also often used, like the discussion page or file galleries, extending the services the system provides.
Consequently, the flexibility that a wiki provides is also a critical point in its usage because it can be like
a Swiss army knife with many different services that can be applied to many different situations. So every
wiki needs a metaphor in order to focus its users to defined ways within which to use the system. The most
popular example is the Wikipedia model that is based on the encyclopedia metaphor to instill quality to
created content.
According to Mader (2008), the Wikipedia model is not the most used model within organizations, and it
is also not the best. This view seems to be a consequence of the work that needs to be done in order to
cultivate a community that cares for the content that is created and actively participates in its evaluation and
maintenance, like the one supporting Wikipedia. The Wikipedian community is the only reason Wikipedia
developed the way it did and keeps on evolving.
Analyzing the wikipatterns community, initiated by the author, where anyone may contribute with what
he calls Wiki Patterns, which may sooth the usage curve of a wiki, and Wiki Anti-Patterns, which may
contradict that goal. We concluded that some of the patterns and anti-patterns he presented are applicable to
organizational wikis using a Wikipedia model, although the author doesnt give much credit to
organizational wikipedias.
In September 2009 project Safirapedia was initiated in a company called Safira, which supplies information
technology consulting services. The main purpose of this project is to have a concept repository with all the
relevant information regarding processes, practices, services, etc.
The project is a part of a plan called Safira 2.0 to turn Safira into a more Enterprise 2.0 organization,
through a series of projects that face different objectives. The projects thought and proposed by Safiras CEO
were: Safirapedia, an enterprise Wikipedia; Google Safira, a google search engine for Safira internal
information; iSafira, a homepage for Safira workers similar to iGoogle; Safira News, a corporate social news
platform; Safirabook, a social network like Facebook for Safira workers; and Knowledge and Collaboration
Platform, which is just what the name suggests.
Using an Action Research methodology, we intended to discover if the use of an organizational
Wikipedia is valuable to a company based on this wiki instance. Through different actions we tried to
achieve practical results for Safira while validating our academic findings, thus, as it usually happens in
Action Research, the researcher was also an actor in the process.
4.1 Contextual Analysis
During the entire life cycle of the project, there were several different analyses to understand what would be
the bigger issues that would arise from the change that was going to happen.
Using Senges Limits to Change perspective we defined the change frontiers: ideologically Safirapedia is
a project that intends to help Safira becoming more close to an Enterprise 2.0; changes were made to
protocols and processes to increase Safirapedians participation; and the methods or tools used are the ones
that will be presented up front in conjunction with the actions taken.
Safiras culture was analyzed through Edgar Schein framework (Schein, 2004) in three levels: (1)
assumptions, the inner desires and needs of the company workers, which were inferred from questionnaires
driven by the researcher; (2) espoused values, the company explicit values; (3) artifacts, the objects that
make up Safira, like chairs, computers or information systems. After this analysis we found that Safira was
indeed prepared for the transformation that had to be done, so we needed to focus on the changes that would
have most impact.
Kurt Lewins Force Field analysis (Lewin K. , 1943) focuses on the forces that support or oppose a change
process. We built a table to help discover what where the main forces influencing participation on
Safirapedia, and it showed that the forces opposing were more powerful than the forces supporting. The main
forces opposing were the lack of time and personal lack of confidence; as for the supporting forces they were
the centralization of knowledge and professional recognition. So we knew that time and confidence were two
ISBN: 978-972-8939-25-0 2010 IADIS
main issues to be improved and centralization of knowledge and professional recognition were two forces
that needed to be motivated and increased.
4.2 Cultivating Change
Using a narrative approach to validate our findings (Figueiredo, 2010) the detail of the description of these
actions is proportional to the level of impact they had on the process. The researcher initiated these actions
while he was searching for results. As the researcher became an integral part of Safira working at the
headquarters, it was based on direct observation and reflection that certain actions were taken.
4.2.1 Selecting the Wiki
In total alignment with the CEO we made a list of the functionalities the wiki needed to accomodate. This
was the first thing to be done in the project and it gave us the parameters that defined our search for the right
platform. They had to be cross-browser compatible, to have robust enough a permissions system, to be easy
to use and to insert contents, to have a space to discuss the articles, to have a friendly interface, to support
several languages, and finally to have integration with Safiras Active Directory. This permanent weekly
alignment with the CEO resulted in a critical analysis of the choices, improving the validation process.
Therefore a comparison was made between several platforms, for instance MediaWiki, Drupal,
Confluence, DokuWiki, that were found using Google search, a web site called Wikimatrix and a Gartner
report (Drakos, Rozwell, Bradley, & Mann, 2009). The choice was made after experimenting the platforms
and comparing, not only the functionalities and interface, but also the price, technical support and longevity.
The platform chosen was Tiki-Wiki because it fits all the requirements listed above, has more built-in
functionalities than all the others, is open-source, and consequently, free of charge, has an active supporting
community and, finally, it has a very friendly configuration panel.
4.2.2 Integrating the Wiki
When integrating a system like this one needs to consider two perspectives: the technical and the social.
On the technical side we integrated it with other systems that already existed, in order to extend
functionalities and avoid turning Safirapedia into an island. The first integration was with Safiras Active
Directory, so we could maintain a concept of single sign on, so workers wouldnt need to know a different
username and password to access Safirapedia, which means less barrieriers. This integration also implies a
more tight security, consequently all logins from outside Safira's internal network were provided through
HTTPS protocol. The second technical integration was with Google Safira, which allows the contents
produced on Safirapedia to be searchable through a Google internal search engine.
On the other hand, the social integration meant that Safirapedia needed to be a part of the workers day-to-
day routine. This was a difficult process: first, there was an initial injection of content that was considered
relevant by the CEO so that Safirapedia wouldnt go online like a blank sheet of paper. This way, the
pressure of being the first content creator and the distance between thinking about writing and actually
writing was diminished.
With a little creativity and some luck with timing, an opportunity appeared in a form of another process
that already existed at Safira the Global Communications Plan (GCP). The GCP was a way to motivate
Safiras different levels of workers to, individually, share their knowledge in different forms. Through written
reports and slideshows, managers, software engineers and software developers could choose what the subject
they would focus on, and every month artifacts had to be created by each function (for instance, managers
and software developers dont create the same reports on a content level).
The partnership created with the GCP was only for software developers. The objective was to make
articles called Programming Tips&Tricks, with a specific focus on a technical issue, and this partnership
made it possible, and motivated the writing of these GCP articles in Safirapedia. Consequently there was a
boost on content creation, with 1 article created per week.
Also on the social side of the integration, we created the Safirapedia Experience to motivate content
creation in group. This initiative was a sort of framework that a group of workers with a common interest and
some shared knowledge could follow. The Safirapedia Experiences had the following conditions:
1. Create a group of 7 people with knowledge about a certain relevant subject;
IADIS International Conference WWW/Internet 2010
2. An expert on the subject must be a part of the group participating in the creation of content or as an
analyst for the content created;
3. Each member must give 1 hour of his week to participate in the Experience and can manage his time
the way that best suits him;
4. Every participant must have information available to connect to the others members.
Each Safirapedia Experience should follow the following steps:
1) Define the subject to talk about and a time span to write it;
2) Get voluntaries to join the group;
3) Organize a list with topics and subtopics. The expert should give his praise to the list;
4) Create teams to write contents. Each participant should choose at least 3 different topics or
subtopics, so teams can emerge from common interests;
5) Create pages.
This framework originated two Experiences: one dedicated to a beginners guide for a Safira technology
named Quartz; and the other dedicated to Project Management at Safira. Unfortunately none of these
accomplished 100% of their initial intent. The beginners guide group created about 40% and the Project
Management group created 0%. Anyhow, there was a good acceptance for the initiative and we believe that,
with practice, people will start adopting the framework.
4.2.3 Promoting the Wiki
Since the beginning of the project Safirapedia that a blog was kept by the researcher under the name of
Safirapedia Blog, with notifications and news about the on going process of implementation of Safirapedia
and with information that would help the readers reflect on their habits regarding knowledge sharing, team
working and collective production. This way we would increase the visibility of the updates that were being
done and contents that were created, motivating people to be a part of the process in their own ways.
Each of the posts done in the blog would then be automatically put on Safiras News that had a voting
system. This way every post could result in a feedback from the users, in terms of liking or disliking what
was written, and pushing others to have a look at the post or referred wiki content. Consequently, there was a
better awareness of the existence of Safirapedia, eventually resulting in more visits.
Promoting the wiki and its usage is a continuum process taking advantage of every opportunity that
appears. In February 2010 there was one more edition of Safiras kickoff meeting to start Safiras new
financial year. This is a social event with some presentations about what was done that year, what would be
done next year and integrates the Safira awards for workers and projects in different areas. There was an
initial intention to do a presentation about Safirapedia and its use, but it didnt happen because there was no
time; nonetheless in a presentation about Accessibility at Safira there was made a huge reference to
Safirapedia, motivating its use. This happened for two reasons: the first, because a wiki is in fact a good tool
to enhance accessibility; the second, because the person that made the presentation was a person with whom
the researcher talked for the most various reasons. This reflects the importance of promotion in every chance
we get, even the slight reference to the tool on a football game, on a water cooler conversation or at lunch can
make a difference.
Another important way of promoting the wiki is doing publicity to the content creators. If they
appreciate that publicity this increases the authors morale and, this motivates them to write more content.
The most asked question was if the page visits were counted and if these values were publicly available, thus
showing the interest in statistical results that authors have, probably to evaluate the impact and visibility of
the content they created.
4.2.4 Supporting the Wiki
The supporting actions that were taken to help Safirapedia were constant along the project and they evolved
with the tool and the community building process.
The researcher provided technical support to the users. When help was needed regarding any functionality
an explanation was given, but if the question indicated something in the tool was going to be a problem, it
triggered a concern to fix it and to improve its effectiveness. These requests for help were dealt through e-
mail or in direct conversations, and usually they were more focused on behavioral attitudes, like how to
correctly format a page or where to link it. The FAQ section on the wiki was also an important part of the
ISBN: 978-972-8939-25-0 2010 IADIS
support given to the community, as not only did it listed the answer to the most common questions asked, but
it also had a functionality that permitted users to suggest new entries.
Guidelines helping the users to interact with each other were also posted, where the process and templates
to create a page on the wiki were detailed, giving the wiki more stability as the creation of new pages was
standardized. This kind of support was much more focused on the communitarian aspects of Safirapedia
because it was intended to solidify the community behaviors and attitudes without imposing rules. As
referred, templates for pages existed so that users didnt need to start from scratch every time they wanted to
write an article. This also prevented the users from worrying about formatting and gave them an idea about
the type of content that was expected.
Regarding another kind of support, in one meeting the CEO confronted the researcher with his low
morale about the project and the results the project was presenting. The results were appearing slowly and in
an unpredictable rhythm justifying the feeling of the CEO, but any disbelief from the top could weaken the
overall projects success. The researcher had to vehemently reaffirm to the CEO, who was the father of the
project, his own confidence on the success of the project, but that this success would only be visible in due
time, as it was an emergent process that was being cultivated, and if rigid time frames were imposed they
could kill the process.
We found in this investigation that a wiki can be useful to an organization. However it has to be accompanied
by a great deal of cultural adaptations and driven by a strong leadership.
Strategic leadership was important for the success of the project. The acts of planning and directing were
done by the CEO when he made Safira 2.0 plan and when he chose someone to take care of it. The project
had strategic leadership and formal support from the administration. The weekly meetings that took place
through the life cycle of the project were a clear sign of interest shown to the project. The CEO was always
present giving advice and revealing information that would, otherwise, never be used.
On another level, the local line leadership integrating the researchers work also made a difference in the
process. First, the fact that the researcher was integrated on the daily Safira life with the workers had a
significant impact on the observation and quality of analysis of the environment. Second, the promotion and
support given to the community has to be a part of any process of implementation of this sort, because people
need to be warned about the existence of new tools that may help their job and on the best way in which they
can use those tools. Although theres a myth about the possible survival of these systems on a pure
communitarian good will, this wasnt verified on this project; if there wasnt a true investment on the project
by the leaders, it wouldnt have enough strength to survive the first months.
The small number of workers of the organization was also a barrier to the process of content creation.
Although the results showed that workers were using Safirapedia the amount of created pages and page views
is always a percentage of the number of users. Safira population is approximately 150 people which means
the number of content creators and content viewers would be limited to a percentage of that. In the course of
this project we only found two people that repeatedly added new content to the wiki, one started willingly to
write about the Information Systems Department processes and the other one wrote because he was a part of
a Safirapedia Experience. All other content that was created by the users was produced either by the
Programming Tips&Tricks initiative or, in a small scale, independent voluntary articles. Quantity is just as
important as quality in this model, because the amount of content creation is directly related to the size and
degree of activity of the community that supports it.
Drakos, N., Rozwell, C., Bradley, A., & Mann, J. (2009). Magic Quadrant for Social Software in the Workplace.
Stamford: Gartner.
Drucker, P. F. (1999). Knowledge-Worker Productivity: The Biggest Challenge. California Management Review , 79-94.
Figueiredo, J. (2010). Actor-networking engineering design, project management and education research: a knowledge
management approach. In P. V. Helander, Knowledge Management. INTECH.
IADIS International Conference WWW/Internet 2010
Hinchcliffe, D. (2007, Outubro 22). The state of Enterprise 2.0. Retrieved Dezembro 2009, from ZDNet:
Lewin, K. (1943, Maio). Defining the 'field at a given time'. Psychological Review , 292-310.
Lewin, K., Lippitt, R., & White, R. (1939). Patterns of aggressive behavior in experimentally created social climates.
Journal of Social Psychology , 271301.
Mader, S. (2008). Wikipatterns. Wiley Publishing.
Mann, R. (1959). A review of the relationship between personality and performance in small groups. Psychological
Bulletin , 241-270.
McAfee, A. P. (2006). Enterprise 2.0: The Dawn of Emergent Collaboration. MIT Sloan Management Review .
Morgan, G. (2006). Images of Organization. California: Sage Publications.
Schein, E. (2004). Organizational Culture and Leadership (3rd ed.). San Francisco: Jossey-Bass.
Senge, P. (1990). The Fifth Discipline: the art and practice of the learning organization. Michigan: Doubleday/Currency.
Senge, P., Kleiner, A., Charlotte, R., Richard, R., George, R., & Brian, S. (1999). Dance of Change. DoubleDay.
ISBN: 978-972-8939-25-0 2010 IADIS
Roberto Pereira*, M. Ceclia C. Baranauskas*, Sergio Roberto P. da Silva**,
Jos Valderlei da Silva** and Filipe Roseiro Cgo**
*Institute of Computing (IC), University of Campinas (UNICAMP), Av. Albert Einstein, 1251, Campinas-SP, Brazil
**Departament of Informatics (DIN), State University of Maring (UEM), Av. Colombo, 5790, Maring PR, Brazil
The folksonomy technique represents an initiative for the organization of the information available on the Web. However,
even systems that apply this technique suffer with problems of lack of quality and information overload when the number
of their users grows. Regardless of these problems, a wide range of systems became available allowing the categorization
of several distinct kinds of objects, but without any level of integration among them. In this paper, we present a critical
analysis about the folksonomy technique, their nature, advantages, problems and challenges. We also propose two
approaches for improving the results obtained with folksonomy. Indeed, we argue that the question now is how and when
assisting users when they are interacting with folksonomy-based systems, in order to make the use of folksonomy more
useful in information organization and retrieval on the Web.
Folksonomy, social tagging, social categorization, folkauthority.
The Web is suffering a quite big evolution, changing from a network that connects webpages, documents,
and resources to a network that connects people, organizations and concepts (da Silva and Pereira, 2008;
Freyne et al., 2007). It is becoming what is now called a Social Web. This new Web is characterized by
facilities for cooperation in contents production and sharing, and by facilities for communication and spread
of ideas among people. Due to the huge amount of contents available on the Web, technical difficulties and
high execution costs make unviable the existence of qualified professionals controlling and evaluating
everything that is produced and published on it. This lack of mechanisms, or measures, to assure the quality
of the information and to organize them, results in a problem now called information overload (Himma,
2007; Levy, 2008). This overload is directly related to the fact that the amount of data, contents and
information is very superior to the humans capacity to absorb, interpret and manage it in a productive way.
The folksonomy technique represents an initiative for helping in the process of organization and
attribution of meaning to the contents available on the Web (Mathes, 2005). In systems that apply
folksonomy, these processes are not restricted to professional editors. These systems adopt the principle that,
if someone is producing and publishing contents, s/he can also be apt to organize and to attribute meaning to
it, labeling it with keywords (tags) in the way they judge more adequate. As a result, folksonomy becomes an
interesting and viable alternative for the open and highly changeable environment that is the Web.
In folksonomy-based systems, the set of tags and objects of a user composes its personomy (da Silva and
da Silva, 2008). This personomy reflects the users vocabularies, preferences, interests, and knowledge.
When the individual personomies are made available and shared among all users of a system, we get the
folksonomy. Thus, a folksonomy is the structure that results from the sharing of all the users personomies
(Mathes, 2005). The sharing and social aspects of folksonomy allow for the interaction among users, what
contributes to the emergence and discovery of interesting knowledge and to the definition of a common
language among them. However, despite these benefits, as the number of users in a folksonomy-based system
increases and, thus, the information being categorized by them also increases, the problem of information
IADIS International Conference WWW/Internet 2010
overload starts to undermine the system usefulness, as in the individual as in the collective (i.e., social) point
of view.
In this paper we are concerned about what should be done for obtaining better results with the application
of the folksonomy technique. We argue that it is necessary to focus on the three pivots of the folksonomy
process (the object, the tag and the user) simultaneously with different views if we want to offer a better
assistance to users of folksonomy-based systems. Also, we should take into account the four tension points in
folksonomy pointed by Smith (2008): i) the individual vs. collective; ii) control vs. freedom; iii) idiosyncratic
vs. standardized and iv) amateurs vs. experts. To achieve this goal, we discuss the folksonomy technique
trying to explain its operation. We give special attention to the four tension points presenting two main
approaches. In the first one, a tag-focused approach, we present a system for centralizing and managing the
users vocabularies and for searching over a set of folksonomy-based systems at once. This system
corresponds to a partial solution to the problem of users having to deal simultaneously with several systems.
In the second one, a users-focused approach, we explain the folkauthority technique, a partial solution for the
problem of the information retrieval quality. Finally, we present some additional minor approaches for
reinforcing our researches and expose our conclusions.
Starting from the middle of 2002, labels of texts grew a lot in popularity as many websites started allowing
their users to accomplish annotations or to use simple words (i.e., tags) in the form of a non-structured text in
a process called tagging (Smith, 2008). Initially these annotations were individual, allowing people to ascribe
some meaning to the website contents by means of classifying its URLs, photos, blogs posts or any other
resource (i.e., object) susceptible to be referenced. The use of this technique creates a personomya set of
tags created by a person to organize resources of his/her interest only. When personomies are made available
and shared among users of a system, we get what is called a folksonomy. Thus, a folksonomy is the structure
that results from the sharing of all the users personomies (Mathes, 2005), i.e., the collective/collaborative
tagging. The whole folksonomy process is essentially based on three pivots (see Figure 1): the userwho
does the categorization, the objectwhich is categorized, and the tagwhich makes the categorization
labeling the object (Mathes, 2005; Riddle, 2005).

Figure 1. The three pivots of Folksonomy.
The usefulness of a tag is very wide (Riddle, 2005). When they are applied to the organization of
information on the Web, tags can aid in the personal information retrieval process, describing and structuring
this information and attributing order and meaning. Among the advantages of allowing users to do the
categorization process are the reductions of costs in time and with investments in specialized services
(Mathes, 2005). To users, the costs are in the cognitive effort to carry out the categorization process and in
the time needed to do it. These costs, however, becomes small when compared to the investment necessary to
maintain an organization done by professionals or specialists, which is impracticable in an open and evolving
environment like the Web.
Russell (2005) argues that folksonomy-based systems allow information to be found on which users
probably would never be presented in another way. Mathes (2005) and Sturtz (2004) characterize this as
serendipity. They assert that folksonomy favors the discovery of knowledge and of useful and interesting
information. Moreover, according to them, perhaps, the most important benefit of the folksonomy technique
is its capacity to reflect the users vocabulary. These systems reflect directly the actions executed by the
ISBN: 978-972-8939-25-0 2010 IADIS
users, their choices, terminologies, i.e., users build and choose the vocabulary that they will use. Mathes
(2005) sustains that, with this non-controlled nature and an organic growth, a folksonomy has the capacity to
adapt very quickly to the changes in the vocabulary and to the users needs, which does not happen in any
other kind of system.
In a folksonomy-based system, there are no formally pre-defined hierarchies for the vocabulary of tags
used in the categorizations. Thus, there is no direct specification of descendants or of a brotherhood
relationship among the terms. The tags applied for the categorization are automatically related by co-
occurrence and group themselves into sets of tags based in common objects. In this way, the folksonomy
process is very different from a formal taxonomy and from other classification schemes in which there are
multiple types of explicit relationships among terms.
One important characteristic of the folksonomy technique, at least from the theoretical point of view, is
the freedom of choice in the quantity and type of words that can be applied to accomplish a categorization.
This process is completely free, in other words, there is no control or scheme to be followed or respected in
such a way that users attribute the tags they want for categorizing an object. This freedom has serious
implications over users cognitive effort to do the organization task, since this effort will be very inferior
when compared to the effort they need to do while selecting a specific classification in a hierarchical
structure like a taxonomy. This is the main reason why such systems are easy to use. Regarding the
information retrieval on the Web, traditional search engines (e.g., Google) uses complex algorithms for the
indexation of Web contents, aiming to present better results for the users queries. Folksonomy-based
systems have the advantage of representing exactly the users opinion, that is, it is not an algorithm
attributing tags to objects, it is the users that are doing thatusers with different opinions and views, in
different contexts and with different goals.
It is important to differentiate two main stages in the process of the folksonomy technique use: the
categorization and the retrieval stages. In the first one, the lack of control results in a great easiness for the
categorization process and provides the benefits previously mentioned. However, the choice of the tags (i.e.,
the category a resource should belong) that best represents an object is not a trivial process. When
categorizing a new object, users will attribute terms they judge more adequate at the categorization time and
in a specific context, but there is no way to foresee or to know if those terms will also be adequate at the time
when users will retrieve the information previously categorized by them. Previous knowledge and even
recent activities accomplished by the users can exercise a lot of influence during the categorization process,
so that users can end up choosing terms that are not frequently used in their vocabulary, that are ambiguous
or that do not have a great meaning in a long time. Moreover, in general, users do not have effective
resources to help them recalling or identifying terms previously used for the categorization of similar objects,
which makes difficult the maintenance of an organized, coherent and well-structured vocabulary. These
questions are stressed when we take into account the existence of several different systems that a user could
interact with.
In the second stage, the information retrieval process suffers with problems of ambiguities, polysemy,
noises and lack of standardization of the terms used for the categorization. These problems are consequences
of the freedom that folksonomy-based systems offer and affect the information retrieval negatively, once
with a chaotic vocabulary, users will have difficulties in finding the content categorized by them before.
These problems are stressed when considered from the collective point of view, since the tags idiosyncrasy
collaborates to found rarities, i.e., for the serendipity, but it also harms the users work in trying to identify an
appropriate term for doing a search. Further, it is necessary to question what the limits of the folksonomy
usefulness are. The chaos originated from the lack of control provides interesting contributions to its easiness
of use and allows the emergency of a bottom-up organization but, currently, it is not possible to determine the
point from which the benefits of this disorganization are neutralized by its problems.
Despite the problems generated by the heterogeneity of users tags, it does not mean the users are wrong
or they are not categorizing information correctly. What users are categorizing is adequate to them or to a
small group, but it will not always bring contributions to the collective or to everyone. There are areas in
which users are more, or less, recommended as information sources and it depends on their capacity, on the
basis and competences on the subjects they are categorizing, i.e., in their cognitive authority (Wilson, 1983;
Pereira and da Silva, 2008)we will explore more on this subject latter in our user-focused approach.
Ohmukai et al. (2005) discuss about the possibility of using a controlled folksonomy. However, as
Mathes (2005) and Riddle (2005) argue, a controlled vocabulary seems to be practically unviable in the case
of applications such as Delicious. On the other hand, several authors (da Silva and Pereira, 2008; Halpin,
IADIS International Conference WWW/Internet 2010
2007; Sturtz, 2004; Wu, 2006) defend the research and development of resources that allow a better use of
the benefits provided by a folksonomy and that could contribute to the quality of the information produced in
folksonomy-based systems. They argue that the main cause of the success of systems like Delicious and
Flickr is the immediate feedback they can provide to users. Moreover, these systems make it visible the
individual and collective aspects related to the users behavior in the information organization process,
something that cannot be obtained with such wealth of details in a system that apply a controlled vocabulary.
The authors show the need of investments in additional resources for the improvement of folksonomy-based
systems, because as we also believe, trying to control a folksonomy is the same as disfiguring it or passing to
adopt a more liberal taxonomy.
In his discussions about tagging, Smith (2008) put focus in four tension points: i) Personal vs. Social, ii)
Idiosyncratic vs. Standard, iii) Freedom vs. Control, and iv) Amateur vs. Expert, that exist in systems that
apply folksonomy.
In Personal vs. Social: the issues here lead to questions like: What is the main goal behind the objects
categorization in tagging-based systems? What is the social nature of tagging? And, what is its great
contribution: selfishness or altruism? Do users make categorizations for their own benefit or are they
motivated by the desire of sharing information, opinions, and knowledge, or to be seen as knowledgeable in a
certain subject, or by other social factors?
In Idiosyncratic vs. Standard: the issues here are related to information organization and retrieval process,
leading to questions like: Should the tags used be unique and idiosyncratic or should they be standardized?
The tags located in the long tail of the users vocabulary are noises or are useful details about users and their
subjectivity? Can these tags be used for disambiguation or context identification, or any other process? Our
tag-focused approach concerns this tension point through the centralization and management of the users
In Freedom vs. Control: the issues here are related to how systems induce users activities, leading to
questions like: Do systems give users complete freedom or do they influence or control the users tags?
When the system offers suggestions, is it changing the behavior that the users would have if no suggestions
were made? Moreover, what can bring more contributions in both information categorization and retrieval: to
control tagging or to keep its freedom nature? Both our approaches (user and tag-focused) keep this freedom
reinforcing our arguments about the need of resources for assist the users instead of controlling the process.
In Amateur vs. Expert: the issues here are related to the quality of the information made available by
these systems leading to questions: How qualified are the people tagging objects? Should the tags created and
the categorizations made by amateurs count as much as the ones made by experts? How does the system
reconcile popular opinion with experts opinions when they disagree? And, How can it be known who are
experts and who are amateurs? Our user-focused approach addresses this question and, together with the
management and centralization of the users vocabulary, aims at improving the quality of the information
retrieved in folksonomy-based systems.
It must be pointed out that all questions raised in this section are emphasized when we consider that users
have to deal with different systems for categorizing different objects. In this context, in the next sections we
present an approach focused on the tags used for categorizing objects, aiming at centralizing and managing
the users vocabularies, and another approach focused on users and directed at the question of the quality of
the information retrieved through the folksonomy technique, preserving its open and evolving nature.
2.1 Managing Personomies
The popularity of the folksonomy technique led to the existence of a wide range of systems. Usually, these
systems focus on a specific kind of object, we can cite as examples: the Youtube, focusing on videos; the
Delicious, focusing on bookmarks; the Flickr focusing on photos; the CiteULike, focusing on papers, etc. The
main goal of these systems is the organization and sharing of their specific objects, in a way that users can
build a personomy and share it with other usersgenerating a folksonomy.
However, once users have created an account in a folksonomy-based system, they do not have any
resource for reusing the vocabulary of tags applied in their categorizations when using another system. It
means that, for each system, users will have to create a different and independent personomy. This situation
contributes to several difficulties and problems in using folksonomy-based systems. To make it clear,
consider the following example: a teacher giving a lecture in a conference desires to categorize and share
ISBN: 978-972-8939-25-0 2010 IADIS
with her/his students the slides of the lecture, the video of the presentation, the pictures that s/he took during
the event, and the blog posts of other participants about the same conference. To do it, the teacher will have
to possess accounts in, at least, four different systems and, consequently, to deal with four different interfaces
and to manage four distinct personomies. We highlight four points regarding this context:
1) The complexity of learning how to use several different systems: As these systems have the objective
of organize resources, it is natural that they have common similarity in terms of features available to the
users. However, the way the interface is designed affects users behavior, expectative and productivity,
because they will have to apply some effort and spend some time in being familiarized with the system and in
understanding it. 2) The categorizations coherence: Because users have to use a different system for doing
the categorization of different objects, it is very difficult for them remembering the tags previously selected
when categorizing objects in other systems. Therefore, the coherence of the tags used to categorize objects
concerning the same context can be compromised. Users can easily forget what tags they ascribed before;
they also can be induced by the system interface (e.g., tags recommendation) or by their previous activities.
3) The management of the users vocabularies: Having different personomies spread over multiple systems
increases the chances of occurring problems in the users vocabularies, such as noises, unmeaning terms,
synonym, polysemy, etc. Furthermore, the management of the vocabulary of tags is a complex activity. If
users need to rename, add, or remove a tag from a set of objects, they have to execute this task manually in
each categorization accomplished with that tag in each system, what is a tedious and easy to fail process. This
situation contributes for the augment of the long tail in the users vocabulary (da Silva and da Silva, 2008).
Tags belonging to the long tail can represent users particularity, knowledge, preferences, etc., but difficulties
in managing users vocabulary harms the benefits that could be obtained from this long tail by inserting
noises terms that will not be useful even to the user who created it. 4) The loss of the knowledge
emergence: A personomy can provide rich information about its creators representing the way they see the
information, how they think about the world and create organizational schemes, what their preferences and
interests are, and so on. However, as the whole is quite different from the sum of its parts, we are losing the
opportunity of studying the users behavior and identifying the users knowledge in a more complete way.
Because their personomies are spread over several systems, we cannot analyze the users when they are
categorizing and retrieving information despite the object and system in use. Consequently, we do not know
if the information obtained when putting all these pieces together would be the same obtained when
analyzing the whole as a unique personomy.
According to our discussions, there are many problems and difficulties in using tags to categorize
information due to its lack of control and structures for the organization, and these problems and difficulties
are emphasized when we consider multiple systems and multiple personomies. The points we explained
above show how this situation harms the maintenance of a consistent personomy and make it difficult for
taking advantages of folksonomy benefits even at the user-level (individual). Following, we present and
discuss an approach for centralizing and managing the users vocabulary and for searching over a set of
tagging-based systems. This approach is being implemented in a system called TagManager.
The TagManager system (TM) is an initiative for allowing users to centralize and manage their vocabularies
and to retrieve information spread over a set of folksonomy-based systems (da Silva and da Silva, 2008). Its
main idea is to provide an environment in which users can manage all their personomies without dealing with
different interfaces and using different vocabularies. Currently, the TM system allows users to synchronize
and manipulate data over four different systems: Delicious, SlideShare, YouTube and Flickr and are available
just for testing. The system aims at reducing the impact caused by the lack of integration among folksonomy-
based systems, which is done by centralizing users vocabulary and by offering resources to organize and
maintain these vocabularies. On the other side, TM also makes it possible to retrieve information over
multiple systems. Thus, when users search for a certain tag, it is possible to retrieve objects categorized with
that tag in each one of these four systems according to the users preference, reducing users workload in
accessing and retrieving information in each one separately.
The important difference between the TM system and other tagging/folksonomy-based systems is its
central pivot. Usually, the central pivot of a tagging-based system is the object being labeled with tags by
IADIS International Conference WWW/Internet 2010
the users. In the TM there is a paradigm shift where the central pivot is the tagi.e., the system centralizes,
organizes and manages tags, regardless of what the objects are and regardless the system in which they were
categorized. Our proposal for the TM system is to build a high-level layer of tags for the users, turning the
management of their tags independent of the folksonomy-based systems and their objects. In agreement with
several authors (Mathes, 2005; Riddle, 2005; Sturtz, 2004), we think that if the users maintain a more
coherent and organized vocabulary, they tend to do a more coherent and organized categorization. Hence,
TM may improve tags usefulness for the organization of personal information, reducing the problems found
when a controlled vocabulary is not applied. Moreover, if users are creating better personomies in the
individual point of view, then a better folksonomy will be obtained when these personomies are shared across
the system, collaborating to the collective/social point of view.
The centralization and management of the users personomies is a keystone for obtaining better results
with the application of the folksonomy technique. Indeed, we argue that it is the first step in developing
resources for assisting users and emerging the social knowledge generated in these systems. Centralizing the
users vocabulary and the personomy makes the reuse, correction and management of tags, as well as the
information recommendation more effectively.
There is another system, called MyTag (, which offers resources for searching
over folksonomy-based systems (for the moment over Flickr, YouTube, Delicious, Connotea, BibSonomy and
BibTex). However, this system has no features for centralizing the users vocabularies and for helping in their
maintenance and organization, which is the main goal of the TM. The focus of the MyTag system is on
searching over other systems (i.e. information retrieval), while TM focuses in the centralization and
management of the users vocabulary and offers information retrieval.
The problem of lack of quality in the information retrieval stage of folksonomy-based systems is partially
because those systems judge with equal importance categorizations done by any user. Consequently, they
will retrieve every object categorized with certain tag by every user that categorized it. There is no way of
filtering or prioritizing categorizations done by people that really understand about the subjects which they
are categorizing (Pereira and da Silva, 2008). The tagging tension point of Amateurs vs. Experts is strong
in folksonomy-based systems due to their intrinsic social nature (i.e., sharing and interaction) and results in a
great amount of irrelevant information whose quality cannot be easily verified.
Wilson (1983) coined the term cognitive authority to explain the kind of authority that influences peoples
thoughts and what they believe. The theory of the cognitive authority put up two important subjects: 1) what
does people resort to second-hand knowledge (coming from others)? And 2) do they resort to whom? The
answer to the first question can be summarized in a word need; and for the second in the following way
people resort to whom they judge to know something that themselves do not know.
Therefore, given that we normally use second-hand knowledge in our lives, it is reasonable to think that
we can also use it to improve the quality of the information retrieved on the Web. We propose to do it using
recommendation about a subject/area from who knows about the subject. Folkauthority (folk + authority) is a
neologism we proposed in (Pereira and da Silva, 2008) designating the authority granted through folksonomy
to the entities that are sources of information on the Web. This concept represents our user-focused approach
for reducing the problems of information overload we explained in previous sections.
To apply the folkauthority concept corresponds to allow users of a folksonomy-based system to ascribe
cognitive authority to the information sources with which the system works. This authority ascription is
accomplished through the attribution of tags that represent the competences, abilities or areas in which
information sources are considered references (i.e., an authority) by their categorizers (i.e., the authority
assigners). Russell (2005) was the first to mention the use of folksonomy for the cognitive authority
ascription with the intention of identifying authorities in a community of users.
The use of the folkauthority process in a folksonomy-based system is adequate for four reasons. First, a
folksonomy is characterized by the freedom of users expression, and that is what users need, since nobody
can say who is a cognitive authority for other person: the person should conclude that. Second, an authority
possesses peculiar characteristics (e.g., sphere of interest, values, etc.) which tags are capable of representing
properly. Third, in folksonomy-based systems categorizations of information are done by their own users,
ISBN: 978-972-8939-25-0 2010 IADIS
this facilitates the process of authority ascription since users are the more recommended ones to evaluate the
categorizers (because they already know them) and are already familiarized with the use of tags and with the
folksonomy technique operation. Fourth, a folksonomy allows to identify more of the context in which the
authority was granted than any other evaluation scheme could provide (a reputation scheme, for instance).
Authorities are not identified based in a single weight/value determined by an algorithm with pre-defined
criteria; instead, they are identified based in peoples subjective opinion, in the diversity and in the
unpredictability of those opinions.
The fact that folksonomy-based systems have their users as categorizers of information makes them a
favorable environment for the application of the folkauthority concept, because the authority ascription starts
to happen over the group of users itself, spreading the authority from its users to their categorizations.
Applying this concept in the TM system is still more suitable, because unifying our two approaches (the tag-
focused and the user-focused) allows recognizing/identifying authorities in different systems, as well as their
informationthe authority ascribed in the TM system worth for all systems with which it synchronizes.
Besides, the authority ascription can also take advantage of the more organized vocabularies that the TM can
provide. We can see a symbiosis relation here, where the folkauthority approach can improve the information
retrieval quality in the TM system, while the benefits generated from the centralization and management of
our tag-focused approach can improve the quality of the generated personomy and by consequence of the
authority definition.
Using the folksonomy technique for the cognitive authority ascription in folksonomy-based systems is
what can be called a meta-categorization. Since, unlike the conventional categorization process in which
users attribute tags describing the content or meaning of an object, in the folkauthority process users attribute
tags to others users of the system (i.e., the source of information) describing their competences, abilities,
knowledge or the areas in which those users are considered as reference. In this way, in a first level there is a
categorization over the objects (information), and in a second level the categorization happens over the
user/sources; a categorization of the categorizers by the categorizers. Therefore, our approach of applying the
folkauthority concept in the TM system represents an initiative of using folksonomy to improve the process
of information organization and retrieval in systems that already apply it.
In this paper, we discuss about the folksonomy technique presenting their main benefits and disadvantages,
and taking into account their two stages of use from the users interaction point of view: the information
categorization and the information retrieval.
Seeing folksonomy through the lenses of the three pivots (the tag, the object and the user) allows
researches and designers to focus their attention in specific challenges while considering the whole dynamic
of tags usage. It is possible to observe this in our two approaches: the TM system and the Folkauthority. The
discussions exposed in this paper about folksonomy become even more critical when they are considered
over multiple systems simultaneously (both challenges and opportunities are emphasized). The approach
implemented by the TM system is focused on the tag pivot (for the centralization and management of the
users vocabulary) aiming at building a better personomy and, consequently, a more structured and organized
folksonomy without controlling the used vocabulary. On the other hand, the folkauthority approach is
focused on the user pivot aiming at identifying users that are cognitive authorities for improving the quality
and relevance in the information retrieval. This approach represents a distinct strategy in which we propose a
meta-categorization that uses the own folksonomy technique to the reduction of problems found in systems
that already apply it.
Both the tag and the user-focused approaches can help in obtaining better results in a
tagging/folksonomy-based system. However, great advantages are only obtained if these approaches are
employed together, once they can create a symbiotic relation in assisting users during the use of a system. For
instance, with the management of users vocabularies we could assist users when they are categorizing
objects as well as ascribing cognitive authority, and with the centralization of users personomies we can
identify every piece of information categorized by the users who are authorities in a certain subject. Hence,
when we apply these two approaches together we are also directing efforts to the object pivot, helping in
questions of information quality, relevance, organization and findability. Indeed, we argue that it is only
IADIS International Conference WWW/Internet 2010
possible to make the use of the tagging/folksonomy technique more effective and more productive if we
consider the three pivots simultaneously.
During the development of our research, we understood folksonomy as a technique that has a lot to
contribute to the information organization on the Web and to social and behavioral studies about users. There
is still a lack of analyses and results about the behavior of folksonomy-based systems and about social
software in general. There are no defined metrics allowing us to affirm whether a social software (especially
the folksonomy-based ones) will obtain success and reach their objective. We are normally induced to such a
conclusion based on the number of users and on the volume of information produced in such systems, but we
cannot say that this is enough. Moreover, there is also a lack of basis for the designers to understand and to
know how to design social software. The complexity of this activity is high due to the users heterogeneity,
and current methods do not support this task efficiently.
As future researches, we see as a challenge the development of approaches and resources for
recommending tags, users and objects. In addition, the development of new schemes of interaction and data
visualization is key for involving and encouraging the users participation, besides allowing them to be more
productive in front the information overload on the Web
The authors would like to express their gratitude to FAPESP (#2009/11888-7), Proesp/CAPES and IC-
UNICAMP for partially supporting this research, and to colleagues from InterHad and GSII.
da Silva, J.V. and da Silva, S.R.P., 2008. Gerenciamento do Vocabulrio das Tags dos Usurios de Sistemas Baseados
em Folksonomia. Proc. of the Workshop de Teses e Dissertaes Webmedia'08, Vitria, ES, Brazil, pp. 201-204.
da Silva, S.R.P. and Pereira, R., 2008. Aspectos da Interao Humano-Computador na Web Social. Proc. of the VIII
Simpsio Brasileiro de Fatores Humanos em Sistemas Computacionais. Porto Alegre, RS, Brazil, pp. 350-351.
Freyne, J. et al., 2007. Collecting Community Wisdom: Integrating Social Search & Social Navigation. Proc. of the
International Conference on Intelligent User Interfaces. Honolulu, Hawaii, USA, pp. 52-61.
Halpin, H., Robu, V. & Sheperd, Hana. The Complex Dynamics of Collaborative Tagging. WWW2007. Alberta.
IW3C2-ACM. 2007.
Himma, K. E., 2007. The Concept of Information Overload: A Preliminary Step in Understanding the Nature of a
Harmful Information-Related Condition. In Ethics and Information Technology, pp. 259-272.
Levy, D. M., 2008. To Grow in Wisdom: Vannevar Bush, Information Overload, and the Life of Leisure. Proc. of the
JCDL05, Denver, CO, USA, pp. 281-286.
Mathes, A., 2005. Folksonomies - Cooperative Classification and Communication through Shared Metadata. Technical
Report available on Folksonomies_Cooperative_Classification.pdf, 2005.
Ohmukai, I., Hamasaki, M. and Takeda, H., 2005. A Proposal of Community-based Folksonomy with RDF Metadata.
Proc. of the Workshop on End User Semantic Web Interaction, Galway, Ireland.
Pereira, R. and da Silva, S.R.P., 2008a. The Use of Cognitive Authority for Information Retrieval in Folksonomy-Based
Systems. Proc. of the International Conference of Web Engineering (ICWE08), New York, USA, pp. 327-331..
Riddle, P., 2005. Tags: What are They Good For? School of Information Technical Report available on, 2005.
Russell, T., 2005. Contextual Authority Tagging: Cognitive Authority Through Folksonomy. Technical Report available
on contextualauthority tagging/conauthtag200505.pdf, 2005.
Smith, G., 2008. Tagging: People-Powered Metadata for the Social Web. New Riders, Berkeley.
Sturtz, D. N., 2004. Communal Categorization: The Folksonomy. Proc. INFO622: Content Representation, 1-8, 2004.
Wilson, P., 1983. Second-hand Knowledge: An Inquiry into Cognitive Authority, Greenwood Press, 1983.
Wu, Harris. Harvesting Social Knowledge from Folksonomies. 17th conference on Hypertext and Hypermedia. 2006.
ISBN: 978-972-8939-25-0 2010 IADIS
Shohei Yokoyama and Hiroshi Ishikawa
Shizuoka University
3-5-1 Johoku Naka-ku Hamamatsu 432-8011, Japan
As described in this paper, we explain how to implement ultra-high-resolution Web applications executed on the tiled
display walls. Although tiled display walls can accommodate contents of various types, they cannot easily handle
standard web contents. Using our new web application design, supporting high-resolution web contents decomposition
for distributed rendering on a tiled display wall. We also describe a design for low-cost tiled display wall using a
"Nettop" computer, and demonstrate our middleware for creating decomposable web applications.
Tiled Display Wall, High-resolution, Web Application, Mashups.
Recently, much high-resolution data exists around us. For example, Google Maps provides various resolution
satellite images of the whole Earth's surface and atmosphere. Flickr provides high-resolution image hosting
for users to share and embed personal photographs. Such services are provided via the Web; users access
them using a web browser.
However, the scope of information that a user can browse depends on the number of pixels of the user's
screen, irrespective of the data resolution. That is, Google Maps' satellite images might display clearly the car
in front of your house, but the neighbor's car is probably off of the screen. On the other hand, if one zooms
out on the map, then both cars become too "small" to see.
Photographs taken using 1 megapixel digital camera have 3680 pixel width and 2735 pixel height (E-3;
Olympus Corp.), but the consumer display monitor resolutions are insufficient to show all pixels of the
photographs. Therefore, we can view only a small fraction of available data.
Generally speaking, for visualization of ultra-high-resolution data, a "tiled display wall" is used. The tiled
display wall is a technique to build a virtual ultra-high-resolution display comprising multiple display
monitors. Many studies have investigated high-resolution display, but the aims of most proposals are for
scientific and medical visualization (Ni et. al., 2006). Therefore, these show high performance but very high
cost. Moreover, standard middleware for tiled display walls does not exist: developers must have deep
knowledge of programming and networking.
As described in this paper, we present a design of a low-cost tiled display wall based only on web
technologies and propose middleware to develop high-resolution web contents to display on the tiled display
wall. The population of web developers is growing at a tremendous pace. Web browsers are used on various
operating systems and involve many standards. Many web services such as Google Maps API and Flickr API
are available nowadays on the internet. We propose a method to use a high-resolution web application that is
executed on a tiled display wall based on the web standards that include HTML, J avaScript, PHP, and so on.
We also present designs of low-cost tiled display walls.
The remainder of this paper is organized as follows. Section 2 presents the design of our tiled display wall
and Section 3 explains our "Decomposable Web Application" concept. Examples of an application are
discussed in Section 4. Middleware for the web-based tiled display wall is also presented in Section 5.
Section 6 describes related works. Finally, Section 7 concludes the paper.
IADIS International Conference WWW/Internet 2010
2.1 Overview
Figure 1 presents our tiled display wall named Wall Display in Mosaic (WDiM). The WDiM system consists
of 16 Full HD (1920 1080) LCD monitors and 16 Nettop PCs(The Player) which have Atom (Intel
Corp.) processors. Consequently, the system is very simple because one PC connects to one LCD monitor. In
addition, WDiM has a game console (The Conductor, Wii; Nintendo Co. Ltd.) as the user interface for web
application on the WDiM, and a web server (The Coordinator) to conduct monitor synchronization
. That is,
WDiM is a distributed model-view-controller model that consists of the Coordinator, the Players, and the
An overview of WDiM is presented in Figure 2. Each Player, Coordinator, and Conductor is connected
via LAN and connects to the internet through the gate-way. Therefore, WDiM can access to the third party
web services and contents on the internet. An external machine, e.g. a cellular telephone, can access to
WDiM via the internet.
The feature of WDiM is in comparison with the existing tiled display walls that the rendering engine is
built on a web browser. The Players display web browser, which is Kiosk Mode. Kiosk Mode is
fundamentally a full screen mode with no toolbars. For example, Internet Explorer in the Kiosk Mode is
available with -k as a command line argument when starting up Internet Explorer.
The Players are fundamentally computers with simple factory settings be-cause a web browser is pre-
installed on a new computer; no extra program is necessary to build WDiM, the tiled display wall system.
The Players must be equal to the number of LCD Panels. Consequently, the installation cost of the
Players might be the single largest component of the WDiM construction cost. For that reason, we must
specifically examine reduction of the installation cost of Players. In fact, no application aside from a web
browser is needed for Players
The Conductor is a web browser too. For this implementation, we choose Wii (Nintendo), which has an
Opera web browser (Opera Software ASA) as Conductor. The Wii is a home video game console; an
intuitive control is available -- the Wii Remote.
The Coordinator is a simple web server; we use Apache2 HTTP daemon on Fedora Linux distribution
(Red Hat Inc.).
2.2 Hardware
The WDiM system consists of low-end consumer-grade computers. Players are Aspire Revo (Acer Inc.)
which has Atom processors (Intel Corp.) and NVIDIA Ion graphics processors (NVIDIA Corp.). The Aspire
Revo performance is worse than that of ordinary computers because the Atom processor (Intel) is designed


Figure 1. Wall Display in Mosaic Figure 2. Overview of WDiM
ISBN: 978-972-8939-25-0 2010 IADIS
for low-end PCs. However, the Aspire Revo is sufficient to browse the Web and process mail. Therefore, it is
called a Nettop computer. Browsing the web is the one and only one task for each Player. For that reason,
we chose Nettop computers as WDiMs Players.
Coordinator is a middle range computer. The specifications for the Coordinator are the following: CPU,
Core2Duo T9300 (Intel Corp.); Memory, 2 GB; HDD, 250 GB/ 5400 rpm.
Generally speaking, tiled display walls such as WDiM are bottlenecked by the performance of the
Coordinator computer because Coordinator must orchestrate all Players. Although network traffic and the
load average are directly related to the number of Players and LCD panels, they do not present an important
problem because the WDiM system described herein has only 16 LCD panels. In addition, much knowledge
and technology of load balancing exists, e.g., web server clustering, web proxy, and so on. The measurement
of scalability remains as a subject for future work.
The four vertical LCD panels and four Player PCs are installed to the same aluminum frames with VESA
LCD monitor arms. Each frame has casters; thereby, WDiM is movable.
As explained herein, WDiM consists of low-cost electronics. We spent less than fifteen thousand dollars
to build WDiM, although it has more than 16 LCDs tiled on a display wall.
2.3 Network
All of the Coordinator, Player, Conductor, and a gateway components connect to the network, which
supports Gigabit Ethernet; the WDiM is connected to the internet via the gateway. Therefore, internal devices
of the WDiM cannot access only the external web services; external devices can also access to the internal
devices of the WDiM. In other words, an iPhone and laptop computer can control a web application on the
In actuality, WDiM, which is based only on web technology, uses no network protocol except HTTP,
which is an application-layer protocol. Consequently, Players cannot send and receive data to each other and
the Coordinator cannot call Players. The web browsers from (Players and the Conductor) send requests to
the web server (Coordinator) and receive responses. This is an HTTP specification.
Consequently, all Players and the Conductor poll the web server on the Coordinator using Ajax behind
web application on WDiM. The flow described above is hidden from WDiM application developers because
WDiM middle-ware provides an overlay network and its access methods over the troublesome HTTP/Ajax
messaging (Figure 3).
Once it was believed that web browsers were useful merely to display static pages that web servers would
provide one after another. However, web pages are no longer static; a good example is Google Maps, which

Figure 3. WDiM messaging Figure 4. Model-View-Controller (MVC) of WDiM
IADIS International Conference WWW/Internet 2010
uses dynamic HTML and Asynchronous J avaScript +XML (Ajax). Using J avaScript, web pages can access
third party web services, download data, and update pages independently Therefore, web applications are
implemented to compose multiple external data resources. In other words, recent data resources, e.g.
RESTful Web services and Web APIs, must be composable.
We return now to a consideration of the tiled display wall. The tiled display wall renders high-resolution
contents in parallel. That is, web applications on WDiM must be decomposable. WDiM is intended to
provide a method that decomposes high-resolution web applications that are implemented to compose
distributed web services and APIs.
Let us consider drawing high-resolution images to multiple web browsers on monitors of the WDiM
system. For example, if the high-resolution image is a 5 MB digital photograph taken by a megapixel digital
camera, then six Full HD monitors are necessary to display every pixel of the image. In this case, at least six
Players send HTTP request and fetch the same 5 MB file. Consequently, 30 MB data are sent for visualizing
a 5 MB image, which is plainly not efficient.
Although multicasting or broadcasting modes that are not supported by web browser might solve this
problem, our challenge in this research is drawing a high-resolution web application based only on web
browsers. Therefore, the Players must each draw a fragment of the high-resolution image on a web browser.
Such a technique is well known. It is a web mapping service application. For example, Google Maps
draws a fragment of the huge satellite image on a web browser. In this case, it is apparent that the image
shown on the web browser consists of many small images (256 pixels256 pixels). Therefore, Google Maps
need not request images that are outside the bounds of the desired image. Actually, WDiM uses such a tiling
method to draw raster images.
A question that remains is whether each Player can draw correct bounds of a big image. The
Coordinator and WDiM middleware help WDiM application developers to resolve this issue. Put simply,
WDiM middleware manages messaging between application logic on the Coordinator and web browsers on
all Players. The WDiM middleware and messaging are described in section 5.
Next, distributed character strings are not supported in this version of WDiM. If a developer does not
declare a bounding box, then strings are sent to all Players whether it draws the string on the monitor or not.
Lastly, drawing a vector image is a subject for future work. We tackle distributed rendering using Adobe
Flash. In this way, distributed rendering on multiple monitors is realized using a combination of existing
To sum up the salient characteristics, The WDiM system consists of a Coordinator(httpd), multiple
Players(Nettop computer), and Conductor(Nintendo Wii). A Coordinator is responsible for the messaging
and application logic, Players draw high-resolution web application to monitors, and the Conductor receive
user interaction. Therefore, the WDiM is logically a kind of Model-View-Controller (MVC) architecture
(Figure 4).
Before describing WDiM middleware, we present two high-resolution web applications of WDiM in this
section to provide an overview of decomposable web application as a mashup of composable web services.
4.1 PhotoMosaic: Using Flickr API
Photomosaic (Silvers R. and Hawley M., 1997 and Finkelstein A. and Range M., 1998) is a technique which
transforms an input image into a rectangular grid of thumbnail images. WDiM PhotoMosaic application
creates Photomosaic images using the large image set of Flickr, the public image hosting service. A
photographic mosaic is a photograph that has been divided into small sections, each of which is replaced with
another photograph having the appropriate average color of the section. While viewing a photographic
mosaic from a distance, it appears as a large photograph, but small photographs can be observed if one
approaches the photographic mosaic. This application aims to demonstrate WDiM, which is not merely a
large display, but also a high-resolution display.
ISBN: 978-972-8939-25-0 2010 IADIS
PhotoMosaic (Figure 5) creates a photographic mosaic consisting of many small photographs from a
photograph that a user gives voluntarily. A large-scale photograph repository is necessary for creation of
photographic mosaic, but PhotoMosaic does not prepare such a repository. PhotoMosaic uses an image
hosting service, Flickr, instead of local photographs, downloading photographs which have average color of a
small section of a given photograph. The PhotoMosaic system is shown in Figure 6. PhotoMosaic uses 75
75 pixels thumbnail images which Flickr generates. No Flickr service exists to search photographs by color;
we extract feature values of color from crawled Flickr images. PhotoMosaic divides the thumbnail into nine
sections, calculates the average color of each section, then saves the feature vector to a metadata database
with thumbnails URL.
An original photograph is given via e-mail by a user along with a notice to PhotoMosaic. The original
photograph is divided and compared in each section with feature vectors of metadata database and get a URL
of closest image. That is to say, PhotoMosaic generates a list of URLs from a given photograph and creates a
photographic mosaic. Then the photograph is appended to list on the Conductors display. The photographic
mosaic that is displayed on WDiM is pointed by Wii remote. A user tilts the Wii remote and scrolls the list of
original photographs, then chooses one photograph from the list. If a user chooses one from list, then the
photographic mosaic is shown in WDiM.
When a photograph is chosen, Coordinator divides the URL list for each Players and sends a subset of
the URLs to corresponding Players. Then each Player downloads thumbnail images in parallel and draws a
corresponding part of the photographic mosaic

See also -umQLok

Figure 5. PhotoMosaic Figure 6. PhotoMosaic System

Figure 7. PhotoMap Figure 8. PhotoMap System
IADIS International Conference WWW/Internet 2010
4.2 PhotoMap: Using Google Maps API
Figure 7 portrays PhotoMap, a WDiM application, which shows high-resolution Google Maps and posts the
geotagged photographs on the map. Geotags are location metadata assigned to the photograph. They
represent a set of latitude and longitude. Geotagging is supported on most cellular telephones and handy GPS
Services showing geotagged images on the map are not novel: Google Maps and Flickr provide such
services. We must emphasize distributed parallel rendering on web browsers on 16 Players. Each Player
draws a different part of the high-resolution map, constituting part of the image of the composite large-scale
The PhotoMap system is portrayed in Figure 8. PhotoMap hosts geotagged photographs; one is shown in
WDiM with a vicinity map of shooting location. A user can change the displayed photograph from the hosted
photographs using a Wii remote and scroll to a new location in which the next photograph is taken coincident
with changing photographs. PhotoMap, in short, is a high-resolution digital photograph frame application.
Registration of photographs to PhotoMap is to upload to Coordinator via FTP and e-mail.
Visualization of earth science data, especially high-resolution satellite images, is a major part of
applications of expensive existing tiled display wall. PhotoMaps shows WDiM, which consists of low-end
consumer machines and which is based only on web technologies, can apply to visualization for earth science.
5.1 Components
In this section, we describe middleware that draws a high-resolution web application with user interaction.
The middleware components are depicted in Figure 9. The middleware consists of a distributed web-
application rendering engine and some third-party J avaScript library components including Ext.J s and
FREDDY (Yokoyama et. al. 2009). The TileCongifuration file includes the number of monitors, size of the
display, and other useful information. WrapperCommander and WrapperReceiver are simple wrapper
programs of PHP; they load TileConfiguration and call Commander and Receiver of a user-developed WDiM
application. The WDiM application consists of Commander, which is executed on the Conductor (Nintendo
Wii/Opera), and Receiver which is executed on the Players. In addition, RESTful Web Services and Ajax
Proxy are placed on Coordinator as necessary.
Commander is an HTML file that includes J avaScript code for Conductor and draws GUI on the monitor,
receives user interaction using Wii Remote and sends commands to Receiver on Players. Receivers has
command handlers, which are defined by the application developer and which await commands from
Commander. When Receivers receive a command from Commander, the corresponding handler is called
with arguments.
We provide abstract classes of Commander and Receiver to application developer. To develop WDiM
applications, the developer writes the subclasses of both Commander and Receiver and extends the abstract
classes to define the application logic.

Figure 9. WDiM middleware components Figure 10. Messaging Mechanism of WDiM
ISBN: 978-972-8939-25-0 2010 IADIS
The WDiM application start-up procedure is the following:
1. Conductor and Players respectively send HTTP requests to the URL of WrapperCommander and
2. Both Wrappers load the TileConfiguration of the WDiM.
3. WrapperCommander sends Commander to Conductor and Wrapper-Receiver sends Receiver to all
4. Commander is a web application displayed on the web browser of the Conductor;
Receivers is a web application displayed on the web browser of the Players.
5. Finally, Commander and distributed Receivers is executed cooperatively and shows high-resolution
web application on the tiled displays of WDiM.
5.2 Messaging
The messaging between Commander and Receiver is based on Ajax over HTTP. However, as described
above, Commander on the Coordinator can not communicate directly to Receiver on the Players because
HTTP does not allow browser-to-browser connection. Therefore, all Players and the Conductor poll the web
server on the Coordinator. Actually, WDiM has messaging of two kinds: Command and SyncData (Figure
10). SyncData is a method used to synchronize data between Commander and all Receivers. If a J SON object
that is declared as SyncData is changed on the Commander, then the object is sent immediately to all Players
to inform Receivers. Command is a method to sending J SON object from Commander to Receivers on
specified Players. The Commander publishes Command and SyncData via methods of WDiM middleware
and the Receiver receives Command and SyncData via a handler for which the user is defined. The WDiM
middleware manages the exchange information between Commander and Receivers.
5.3 Tile Configuration
Commander and Receiver get the TileConguration when starting a WDiM application. Figure 11 shows
properties for our 16-monitor WDiM tiled display wall. User only defines the width and height of LCD
monitor and its bezel and Number of LCD monitors of the Tiled Display Wall. In addition, some useful
properties, e.g. distance from the center and bounds of the LCD monitor draw, calculated. The WDiM
application can use the properties to draw the high-resolution contents.
The WDiM middleware and applications can correspond to various sizes of tiled display walls by
changing the TileConguration. For example, it is possible to build a tiled display wall made of two laptop
computers arranged side-by-side on the desk.

Figure 11. Tile configuration
IADIS International Conference WWW/Internet 2010
Many proposed approaches are used for building tiled display wall systems. NASAs Hyperwall (Sandstrom
et. al., 2003) has a 64-megapixel Tiled Display Wall comprising 49 monitors (7 monitors horizontal 7
monitors vertical). LambdaVision uses 55 monitors (11 monitors horizontal 5 monitors vertical) and builds
a 100 megapixel high-resolution tiled display wall system. They also propose middleware for tiled display
walls called SAGE (Renambot et. al., 2003) An extremely large tiled display wall is HIPerSpace (DeFanti 2009) used at the University of California, San Diego. Altogether, it has a 225-megapixel display.
Details of existing tiled display wall systems were surveyed by Ni et. al.(2006).
Those systems aim at performance and resolution for application to scientific visualization for life science
data and ultra-high-resolution satellite images. Apparently, research issues related to tiled display wall have
bypassed consumer applications in favor of improving scientific high-performance computing. Consequently,
such applications require the use of expensive high-end machines, complex settings and elaborate
programming. However, in explosive growth of web technologies, ultra-high-resolution satellite images, e.g.
Google Maps, via web browser are becoming increasingly available, and are valued by many users.
In other words, high-resolution visualization is no longer for scientists only: ordinary people can have
access to it. For that reason, we propose a low-cost tiled display wall, WDiM, which consists of low-end
machines and which is based solely on web technologies.
Additionally, we must point out that WDiM uses only web programming language. Numerous skillful
web developers and programmers are working today who can create WDiM applications.
As described in this paper, we propose WDiM, a method to display high-resolution web applications on a
tiled display wall. We also express the design of low-cost tiled display wall made of low-end consumer
devices. Additionally, we presented an example of a high-resolution web application which renders to the
tiled display wall. This paper explained how to decompose high-resolution web applications implemented to
compose distributed web services and APIs.
Future research in this area will be undertaken to attempt browser-to-browser communication using a new
version of the Opera browser and vector rendering using HTML5. We believe that decomposable high-
resolution web applications are a novel model of next-generation web usage.
DeFanti T. et al, 2009. The optip1ortal, a scalable visualization, storage, and computing interface device for the optiputer.
Future Generation Computer Systems, The International Journal of Grid Computing and eScience. 25(2), pp.114-123.
Finkelstein A. and Range M., 1998. Image Mosaics. In Roger D. Hersch, Jacquesl Andr, and Heather Brown, Ed.,
Artistic Imaging and Digital Typography. LNCS, No. 1375, Heidelberg: Springer-Verlag.
Ni T. et al, 2006. A survey on large high-resolution display technologies, techniques, and applications. In Proceedings of
Virtual Reality Conference. pp. 223-236.
Renambot L. et al, 2004. Sage: the scalable adaptive graphics environment. In Workshop on AdvancedCollaborative
Environments (WACE04).
Sandstrom T. A. et al, 2003. The hyperwall. In Proceedings of International Conference on Coordinated and Multiple
Views in Exploratory Visualization. pp. 124-133.
Silvers R. and Hawley M., 1997. Photomosaics. Henry Holt, New York.
Yokoyama S. et al, 2009. Freddy: A web browser-friendly lightweight data-interchange method suitable for composing
continuous data streams. In Proceedings of First International Workshop on Lightweight Integration on the Web
(ComposableWeb'09). In conjunction with ICWE 2009, pp. 39-50.
ISBN: 978-972-8939-25-0 2010 IADIS

Alssio Miranda J nior and Laura S. Garca
C3SL, Computer Department, Federal University of Paran UFPR, Curitiba-Pr, Brazil
In a time when the most successful development efforts in Machines Translation (MT) are based on closed software,
Apertiumhas become an alternative mature, interesting and open source. However, one of the main obstacles for the
improvement of its results and popularization is the absence of a specific interface to manage its linguistic knowledge,
which, because of the imposed difficulty, reduces the number of potential collaborators to the development of the
language pairs. In the present paper, we propose an interaction-interface environment that can abstract the concepts of the
system and the textual process available for the development of bi or monolingual dictionaries. In addition to that, it has
the ability to capture and organize knowledge in a simple manner for non-experts in Computing, leading to the growth
and development of new language pairs for the MT.
Apertium, Machine Translation Systems, WiKLaTS, interfaces, knowledge management, WiKLang.
Machine Translation (MT) (Hutchins, 1992) is the traditional term used to refer to the semi or fully-
automated process whereby a text or utterance in a natural language (so-called source-language) is translated
into another natural language (so-called target-language), resulting in an intelligible text which, in turn,
preserves certain features of the source-text, such as style, cohesion and meaning.
In recent years, we can see that the successful instances of Machines Translation (MT) are always a set of
closed software and knowledge base, distributed as static products and with a commercial purpose. This
model comes with a big disadvantage, namely the difficulty imposed to the architecture and technique
improvement studies and even to the development of language pairs without a financial incentive.
This situation hinders the development of open-source software able to transform this technology in
something open and democratic. In this sense, creating opportunities for the community to contribute even
more to the evolution and the development of this interdisciplinary area that involves Computer Science and
Linguistic has become a relevant challenge.
Among the different types of MT software that have appeared following the open-source software model
and focusing on the abovementioned objectives, we have chosen the Opentrad Apertium Project
(Apertium) (Forcada, 2008) as a consolidated free/open-source MT platform which is in constant evolution
and with expressive results in literature (Tyers, 2009 and Forcada, 2006).
Apertium is a shallow-transfer Machine Translator based on superficial syntactic rules that use its own
knowledge (data) base, with an open and flexible structure in XML standard. It provides an engine and
toolbox that allow users to build their own machine translation systems by writing only the data. The data
consists, on a basic level, of three dictionaries and transfer rules.
Shallow transfer systems, such as Apertium, based on superficial syntactic rules, use superficial
information based on the structure of the sentence, avoiding the in-depth semantic details within that
structure. In general, this type of MT is more efficient and can substitute a complete syntactic analysis with
satisfactory results.
IADIS International Conference WWW/Internet 2010
Although Apertium has technical advantages widely recognized and praised by the international scientific
community, and despite its structured and open knowledge base, its popularization has been facing a number
of obstacles. The development of language pairs for Apertium is perceived as complex, based on lengthy
XML files without interface alternatives that allow for the development and maintenance of linguistic
knowledge bases. Therefore, all efforts are based on textual tools.
This fact is a hindrance to the evolution of the available language pairs precisely because the number of
users able to develop knowledge for the tool is strictly limited to experts and/or trained users.
When new users are interested in contributing to the development of a given language pair, they must go
through a period of identification and learning about structures, command line tools and processes, which
might be extremely complex for a layman in Computing. Indeed, this learning period may become rather
long and complicated due to the lack of formalization and tools, thus discouraging countless potential
collaborators. In this light, the main objective of the present work is to minimize this negative factor.
Among the aspects that must be lightened and simplified within the interaction-interface environment, we
would like to mention the following: the marking of XML files, the compilation of XML bases in Apertium,
as well as the organization, the acronyms and the internal structure of the knowledge base. Furthermore, with
the ever-growing knowledge about language pairs, the direct manipulation of these files has also become
intricate from a human point of view.
In the present paper, we present a formalization proposal of a part of the process of a language pair
creation and development for translation, which we describe through interfaces and stages of knowledge
specification. This study has allowed for the dissemination of a set of interfaces which, in turn, make up a
more adequate and efficient alternative to the current process. We named this set WiKLang, and its main
objectives are the following:
Standardize both the understanding and the process, and make them less dependent on textual
Reduce to a minimum the Computing knowledge needed for a new collaborator to understand how
to create and develop language pairs;
Allow for the manipulation of the knowledge bases in a reliable, generic fashion, including
management functions;
Stimulate both translation improvement between existing language pairs and the development of
new pairs, thus making MTs, such as Apertium, more reliable and with a bulkier knowledge base.
In order to achieve the previously mentioned objectives, we have followed the steps described below:
Interaction with the Apertium team so as to understand the interface and its architecture;
Identification of problems and aspects that could be improved in the interaction with Apertium
Formalization of the knowledge specification process used in Apertium;
Creation of a project of an interaction-interface environment (WiKLang) using concepts similar to
Apertiums, as a more adequate alternative to the linguistic knowledge specification needed to translate
language pairs;
Development of a non-functional prototype to be assessed by the Apertium team.
Apertium is an open-source platform for creating rule-based machine translation systems. It was initially
designed for closely-related languages, but has also been adapted to work better for less related languages.
ISBN: 978-972-8939-25-0 2010 IADIS
The original shallow-transfer Apertium system consists of a de-formatter, a morphological analyzer, a
categorical disambiguator, a structural and lexical transfer module, a morphological generator, a post
generator and a reformatter (Tyers, 2009).
Inserting a new language pair to be translated by Apertium is a challenging task that requires advanced
knowledge of Linguistics in addition to specific Computing knowledge. A researcher or user that may wish
to develop translation pairs for Apertium relies on non-specific tools and very few interface resources and
scripts, which in turn are fully textual and detail parts of the process only.
The architecture of Apertium consists of a generic translation machine that may be used in association
with different knowledge bases, such as dictionaries and rules of shallow-transfer between languages. It is the
combination of these elements that does the translation of text through a process of shallow-transfer, whose
main distinctive feature is the fact that it does not carry out a full syntactic analysis, but rather operates over
lexical units.
Therefore, one must master certain key concepts so as to understand both the knowledge model and the
basic structure of the XML files to be created.
Some of these key concepts are lexemes, symbols and paradigms. A lexeme is the fundamental unit of
the lexicon of a language (word). A symbol refers to a classifying tag, which may be either grammatical or
linked to the execution stages of the machine. A paradigm is a script containing morphological flexions for
a given group of words.
Moreover, there are three stages one must go through in order to insert new language pairs into Apertium,
stages which, in turn, give rise to XML files. It is important to stress that the entire process is based on
textual files, and that the references among the tags, for example, must be kept correctly and manually
Table 1 shows these three defining stages, namely the development of a monolingual dictionary, a
bilingual dictionary and translation rules. We have developed solutions for the first two stages.
In order to create and manage the files that make up the dictionaries and rules, one must necessarily know
and understand the syntax, the labels, the internal structure and the tagging hierarchy. Typical monolingual
dictionaries, for example, currently comprise more than 10 thousand lexemes spread through 150 thousand
lines of XML information, which makes knowledge management a task too complex to be carried out
manually and without specific techniques and support tools.
In order to make terminology clear, in the present paper we will refer to fixed XML Apertium tags as
elements, leaving the concept of tag to flexible language features. We will present below examples of
excerpts of Apertium files, including the meanings of the elements mentioned whenever necessary. The
information presented here is largely based on tutorials provided by the platform.
Table 1. Data for the apertium platform: development stages
Development Stage Content
Monolingual Dictionary of language xx and yy Contains the rules of how words in language are inflected.
Bilingual Dictionary for each direction xx to yy
and yy to xx
Contains correspondences between words and symbols in the two
languages in the direction defined.
Transfer Rules for each direction Contains rules for how language xx will be changed into language
4.1 Dictionary Format
Both monolingual and bilingual dictionaries share the same XML specifications. We shall now present an
overview of the standard structure of these files.
The element dictionary comprises the entire content of the file and is basically divided into four parts
isolated by tags and their respective content, as follows:
alphabet: set of characters used to perform tokenization;
sdefs: markings that can be applied to structures during the process;
pardefs: flexion paradigms applied to the lexemes;
section: lexeme definition and special symbols belonging to the language, possibly involving
Example of the Basic Schema of an XML Dictionary:
IADIS International Conference WWW/Internet 2010
<?xml ver si on=" 1. 0" >
<di ct i onar y>
<al phabet ><! Char act er s used i n t he di ct i onar y - - ></ al phabet >
<sdef s><! - - Symbol s t hat can be appl i ed dur i ng t he pr ocess; - - ></ sdef s>
<par def s><! - - Fl exi on par adi gms appl i ed t o t he l exemes; - - ></ par def s>
<sect i on i d=" mai n" t ype=" st andar d" >
<! - Sect i ons l exeme def i ni t i on and speci al symbol s
bel ongi ng t o t he l anguage, possi bl y i nvol vi ng par adi gms. - - >
</ sect i on>
</ di ct i onar y>
4.2 Monolingual Dictionary
Monolingual dictionaries contain lexemes, flexions and all other elements needed to morphologically classify
a sentence, expression or word in the language in question. There may even be ambiguities in this
classification, but they are cleared in subsequent translation steps, which we will not discuss here. The
platform works with parts of the language, hence the need to work with two morphological dictionaries in
order to carry out translations.
Say we wish to translate the noun livro, in Portuguese, which corresponds to book in English. In
order to do so, we need to create a monolingual dictionary in English as well as one in Portuguese. We will
now describe the development of a small monolingual Portuguese dictionary. Figure 1 shows the result of the
analysis process of the word livro.

Figure 1. Analysis process of the word livro.
The first stage is the definition of the alphabet, i.e. the set of characters used in given language (expressed
by the equally named element alphabet).
Example of alphabet tag
bcdef ghi j kl mnopqr st uvwxyz</ al phabet >
After that, all symbols that may be used in the translation process must be registered, and they may either
have grammatical semantics or be part of the internal treatment of the process. The element sdefs
comprises this content.
Example of sdefs tag.
<sdef s>
<sdef n="subs" c=Noum/ > <sdef n=" adj " c=Adj et i v/ >
<sdef n="sg" c=Si ngl e/ > <sdef n=" pl " c=Pl ur al / >
<sdef n="m" c=Masc/ > <sdef n=" f " c=Femal e/ >
<sdef n="mf " c=Undef Gender / >
</ sdef s>
Next, one must define a section of paradigms comprised by the element pardefs, to which the paradigm
used by nouns as livro is added. In Portuguese, the lexeme for nouns is always male and singular. The
singular form of this noun is livro and the plural is livros.
Example of paradefs tag:
<par def n="l i vr o__subs">
<l / >
<r ><s n=" subs" / ><s n="sg"/ ></ r >
</ p></ e>
<l >s</ l >
ISBN: 978-972-8939-25-0 2010 IADIS
<r ><s n=" subs/ ><s n="pl "/ ></ r >
</ p></ e>
</ par def >
Table 3 shows a few XML elements used in Apertium. The attribute n of the <pardef>element simply
defines a name of reference for the paradigm. In order to make its identification easier, this name follows a
terminological standard first defined by the Apertium pioneers.
Within the pair of directions, knowledge is built from left to right, conceptually representing the direction
of the generation process. This is necessary because the knowledge bases, dictionaries and rules are compiled
for a finite machine. In the case of a dictionary, the movement from left to right corresponds to the process of
word analysis, whereas the movement from right to left corresponds to the construction of words, as shown
in Table 2.
Table 2. Description of XML pardef process
In Direction Out Process
livros LefttoRight Livros<subs><pl> Analysis
livros<subs><pl> RighttoLeft Livro Generation

The element that succeeds paradefs, entitled section, refers to sections of lexical content. An
important attribute of <section>is type, which in turn defines the kind of section and may be classified as
standard (indicating that the content consists basically of lexemes) or unconditional (indicating
punctuation and other symbols we will not discuss in the present paper).
Table 3. Description of XML elements
Element Description Element Description
E EntryineitheraParadigmoraSection LM Lema
P PairofaElement I Radicaltobeused
L LeftDirection PAR ParadigmName

The link between lexemes and their respective (previously defined) paradigm is established within the
element section. Table 3 shows the description of the XML elements.
Below is the definition of the lexeme livro:
<e l m=" l i vr o" >
<i >l i vr o</ i ><par n="l i vr o__subs"/ >
</ e>
In order to show how the basic structure of a monolingual dictionary works, we have developed this very
simple version of a Portuguese dictionary using the word livro.
4.3 Bilingual Dictionary
Bilingual dictionaries make correspondences between lexemes and symbols of both languages involved, with
their content being the knowledge needed for translation in both ways. Indeed, their main content consists of
the word-for-word relation of correspondence between the languages in question.
The structure of a bilingual dictionary is the same as that of a monolingual dictionary, but its objective is
to register and map lexemes across two different monolingual dictionaries. Nevertheless, even though the two
kinds of dictionaries share the same structure, their descriptive forms are different.
The alphabet element is unused in bilingual dictionaries: as mentioned earlier, it's only used for
The element sdef must contain the union of the elements present in the two monolingual dictionaries
involved. The element pardef may contain advanced paradigms for bilingual definition, and therefore
cannot be applied to our example.
The element section is the one to hold the relations between lexemes, and its function is to manage the
standard exchange between words. These standards play the role of the word-for-word translator, though
with a more thorough analysis of the results. When there are ambiguities (such as the definition of undefined
IADIS International Conference WWW/Internet 2010
genders), different resources are used in different translation stages which, in turn, we will not discuss in
the present paper.
Example Fragment of word mapping in a bilingual dictionary.
<sect i on i d=" mai n"t ype="st andar d">
<l >l i vr o<s n=subs" / ></ l >
<r >book<s n=subs"/ ></ r >
</ p></ e>
</ sect i on>
4.4 Translation Process
Figure 2 depicts the translation process of the source word livro, in Portuguese, and its correspondent in
English, book. Firstly, the input element is analyzed taking the respective monolingual dictionary into
account. Next, possible manipulations of the translation rules may take place. Finally, the bilingual dictionary
generates the new concept in the target language.

Figure 2. Simple translation process.
The main objective of WiKLang is to import and export the XML files of monolingual and bilingual
dictionaries, allowing for their management to take place through an interaction-interface environment
which, in turn, makes the process more objective and transparent. This way, our aim is to offer an alternative
to the current Apertium process of file creation in the stages of mono and bilingual dictionary development.
The fact that the proposed interface must be generic enough to allow for the creation of mono and bilingual
dictionaries in any pair of languages is what we find particularly challenging in our project.
The strategy we adopted in the present study consisted of the development of a prototype, which was then
submitted to the assessment of potential users. Finally, we analyzed the feedback obtained, aiming at
improving the environment. Indeed, it was an interactive process of assessment and improvement. We would
like to mention that, even though we have paid special attention to the requirements of interface design, we
did not begin the initial stages of prototype development by revising HCI literature.
5.1 The Environment
Figure 4 shows the five main elements of the interface, as follows:
1. A tab bar displaying the three functions, namely monolingual dictionary, bilingual dictionary and
transfer rules;
2. A context bar showing important information about the current status of the process underway;
3. A bar showing the contextualized sub-processes according to the process chosen;
4. An action panel, or the main work point of the user, where the actions of system orientation and
information input takes place;
5. An instruction panel, or guidelines to support users during the process.
ISBN: 978-972-8939-25-0 2010 IADIS

Figure 4. WikLang Statistics and information
5.2 Monolingual Dictionary
We divided the development of the monolingual dictionary into four sub-processes, which in turn refer to the
structural distribution of the resulting XML files.
The first of such sub-processes is the definition of the monolingual dictionary in terms of basic attributes
and statistics. Attributes such as the universal language code, description and version are merely informative,
but the alphabet, on the other hand, influences the translator platform. The environment also counts on basic
statistics to help control the development of knowledge about the language (see Figure 4).
5.3 Symbol Definition
The symbols must manage all tags that may be used to identify lexemes. In Apertium there is no hierarchy
among the tags of monolingual dictionaries. However, such hierarchical classification is an important tool for
the linguist-user. Therefore, we have decided to include this kind of symbol classification even though it is
not used by the MT, being useful only for the orientation in knowledge construction.
For each symbol, we propose a code definition, as well as a description in the source language. In
addition to that, we present statistics about the relation among the symbols in the category in question.
Therefore, the interface allows for the hierarchical association among symbols based on the grammatical
classification or concept groups (see Figure 5).
5.4 Paradigm Definition
The definition of paradigms is certainly one of the most complex parts of the management process of
bilingual dictionaries.
Users may either choose an existing paradigm or create a new one. Next, they must name the paradigm
and choose an example word having its radical highlighted, using the standard Apertium terminology.
This sub-process was subdivided into two smaller stages. The objective of the first is analysis, displaying
all flexions provided by the paradigm and interacting with dynamically inserted lexemes, so that users can
have a good overview of their behavior and flexions.
The second stage, on the other hand, aims at editing elements of the paradigm. Among the support tools,
we have an expression editor that may take the right-hand side of the element, as well as word definitions and
symbol chains that define the classification in question. These chains, in turn, are built through drag and
drop resources in searchable lists, eliminating typos and making rules clearer (Figure 6).

Figure 5. Symbol Definition Interface Figure 6. Paradigmdefinition
IADIS International Conference WWW/Internet 2010
5.5 Lexeme Definition
One of the most common tasks in lexeme definition consists of consistently connecting lexemes to the
paradigms that generate their flexions. In order to make this process easier, we have created a wizard to
locate lexemes and establish these connections, as well as to choose the most adequate paradigm.
To define a lexeme, users must enter a word and highlight its radical, and provide four example flexions.
An initial analysis phase checks the available paradigms and suggests those that suit the flexions given. It is
then up to users to choose the most adequate options, and they are also free to complement or create new
paradigms. For those lexemes that have a specific flexion, such as compound nouns, one can create an
embedded paradigm, exclusive to that lexeme (Figure 7).
5.6 Bilingual Dictionary
We divided the present stage into two sub-processes, namely: definition and management of correspondences
between lexemes and expressions.
In the first sub-process, users select two monolingual dictionaries to be related, and the system prepares
the definitions of the bilingual dictionary automatically, with the alphabet and a grid of symbols derived from
the relation between the languages. Additionally, both statistics about the monolingual dictionaries and
information about the relations are available.
A grid of the unified symbols displays the following information in columns: the code, the description
and the morphological dictionary to which the symbol belongs. These pieces of information are important for
the orientation within the interface, and must also be included in the resulting XML file.
The second sub-process manages the relations between lexemes and expressions, and indentifies the
relations that truly define the bilingual dictionary. Building these relations is a rather peculiar task, since the
process takes place on lexeme level, simulating relations word by word; in other words, the system builds
relations based on the partial interlingua generated.
Figure 8 shows the interface and the existing relations for a given lexeme. In the lateral bar, users may
choose options to filter and search definitions related to a certain lexeme both in the A language (i.e. the
source language) and in the B language (i.e. the target language). Once they have chosen a lexeme, they
have two grids with information about it.

Figure 7. Lexeme definitions

Figure 8. Bilingual dictionary and the relations between
lexemes and expressions
The first of such grids contains all the possible flexions of the lexeme chosen, as well as its
correspondences from the analysis phase based on the morphological dictionary. The second grid, on the
other hand, displays the rules associated to the lexeme, which in turn are organized in three columns, as
follows: the first column displays the standard accepted by rule; the second displays the output to be
generated; and the third shows the valid orientation for that rule.
The definition displayed in the first column follows the principle of pattern matching with the most
specific rule or the most complete standard. It refers to the root of the lexeme, followed by its morphological
classification generated by analysis.
The content of the second column will substitute the standard found, so as to prepare the phase of
generation into the target language. Finally, the third column displays the direction into which the rule is
valid considering that certain rules are valid solely in one direction within the language pair.
ISBN: 978-972-8939-25-0 2010 IADIS
We believe that the Apertium platform is an important open-source initiative within MT; nevertheless, it still
needs bulkier tools to support it in its ambitious task of matching traditional commercial software.
By analyzing one part of the process, namely the development of both mono and bilingual dictionaries,
we noticed that there is a lot of difficulty due to the lack of tools to assist in knowledge input and
management. For this reason, we proposed an interaction-interface environment, named WikLang, to make
this process of knowledge specification easier.
We feel that the WikLang interface and cycles proposed here indeed help in the development of larger,
bulkier dictionaries for the Apertium platform. Without special systems as the one introduced here and from
a strictly human point of view, this task would be rather difficult
As far as future works are concerned, we would like to suggest the following:
A thorough analysis of the HCI concepts related to the environment;
The implementation of new requirements pointed out by users after their first trial of the functional
The extension of the interface so as to contemplate relations between translation rules as well;
The analysis of the generalities of the interfaces, so that they allow for the development of any
The reflection about other modules to be integrated seeing as WikLang is the first implemented
module from the WiKLaTS project.
This work was supported by National Council for Scientific and Technological Development (CNPq/Brazil)
and National Council for the Improvement of Higher Education (CAPES/Brazil).
Forcada, M. L. 2006. Open-source machine translation: an opportunity for minor languages. Strategies for developing
machine translation for minority languages. 5th SALTMIL workshop on Minority Languages.pp. 17
Forcada, M. L. and B. Ivanov Bonev and S. Ortiz Rojas and J. A. Prez Ortiz and G. Ramrez Snchez and F. Snchez
Martnez and C. Armentano-Oller and M. A. Montava and F. M. Tyers. 2008. Documentation of the Open- Source
Shallow-Transfer Machine Translation Platform Apertium.
Hutchins, W. J ohn, and Harold L. Somers. 1992. An Introduction to Machine Translation. London, Academic Press.
ISBN 0-12-362830-X.
Ramrez-Snchez, G., F. Snchez-Martnez, S. Ortiz-Rojas, J . A. Prez-Ortiz, M. L. Forcada, 2006. "Opentrad Apertium
open-source machine translation system: an opportunity for business and research", in Proceedings of Translating
and the Computer 28 Conference, London.
Tyers, F. M., L. Wiechetek and T. Trosterud. 2009 "Developing prototypes for machine translation between two Smi
languages". Proceedings of the 13th Annual Conference of the European Association of Machine Translation,
IADIS International Conference WWW/Internet 2010
Mrs. Ghada Aldehim
, Dr. Pam Mayhew
and Mr. Majed Alshamari

School of Computing Sciences in the University of East Anglia - Norwich- United Kingdom
Recently, the growth of the Internet has led to an explosion of website content, and therefore research into usability is
extremely important. The heuristic evaluation method is one of the usability inspection methods that is employed to
evaluate Internet websites in order to assess and improve their usability. However, heuristic evaluation suffers from a
number of drawbacks; therefore, we are exploring a customised guidelines method designed especially for e-shops. This
method is compared with the traditional heuristic evaluation method by employing both methods to compare which
method is able to offer more efficient support to the evaluators, and hence to generate effective feedback. In order to
achieve this aim, two electronic shops are chosen and used to carry out the evaluation process. Eight expert evaluators are
engaged in this study, divided into two groups of four. Each group evaluates two electronic websites. The analysis and
findings of this study clearly show that the customised guidelines method is able to offer better results than the traditional
heuristic evaluation method.
Heuristic evaluation, evaluators, usability, customised guidelines, website evaluation.
Researchers need to focus on the usability of Internet websites as it is one of the fastest-growing areas. This
can be clearly reasoned because of a number of critical factors such as the number of the Internet users, the
number of websites, the efforts of designers and developers, and the services offered via the Internet (Scott
Henninger, 2001). Several evaluation methods can be applied and used in order to measure website usability;
one of these methods is the inspection method, which includes: heuristic evaluation, cognitive walkthrough,
and feature inspection (Jacobsen Nielsen, 1992). Among these techniques, heuristic evaluation is
recommended as the fastest, cheapest, most flexible and effective way for identifying usability problems in a
user interface (Saul Greenberg et al., 2000). This form of evaluation is a method for finding usability
problems in a user interface design by having a small set of evaluators examine the interface and judge its
compliance with recognized usability principles (the 'heuristics') (Jakob Nielsen and Rolf Molich, 1990).
This paper starts by exploring the current literature on heuristic evaluation, its pros and cons, and then how
the customised guidelines method was derived. This is followed by the papers objectives and the approach
taken in order to achieve the papers aim. It concludes with data analysis.
This section discusses the heuristic evaluation method, how it is employed and the number of evaluators
needed. It also explores its advantages and disadvantages. It then presents the need for a customised method
and how it was proposed.

ISBN: 978-972-8939-25-0 2010 IADIS
2.1 Heuristic Evaluation Method
Heuristic evaluation involves having a small set of evaluators examine an interface and judge its compliance
with recognized usability principles (the 'heuristics'). This method was developed by Nielsen and his
colleagues in 1990, guided by a set of usability principles, to evaluate whether a user interface conforms to
these principles (Jacobsen Nielsen, 1992) (Hix et al., 2004). These principles or criteria are referred to as
guidelines or general rules that can guide a design decision or be used to critique a decision that has already
been made. In general, heuristic evaluation is difficult for a single individual to do because one person will
never be able to find all the usability problems in an interface. Moreover, many different projects have shown
that different people find different usability problems (Morten Hertzum and Nielsen Jacobsen, 2001). As a
result, to improve the effectiveness of the method significantly, it should involve multiple evaluators.
According to a number of researchers, such as Jakob Nielsen, a single evaluator is able to identify only 35%
of all the usability problems in an interface, which is the result of averaging more than six of his projects.
However, since different evaluators tend to find different problems, it is possible to achieve a better
performance by increasing the number of evaluators. Figure 1, clearly shows that there is profit from
involving many evaluators. Nevertheless, there is no agreement about the number of evaluators that are
enough to identify a sufficient number of usability problems. According to Jakob Nielsen (1994), three to
five evaluators typically identify about 75-80% of all usability problems. Koyani (2003) asserted that two to
five is enough.

Figure 1. The relationship between the number of evaluators and the number of the found problems (Jakob Nielsen,
2.2 Advantages and Disadvantages of Heuristic Evaluation
Heuristic evaluation is recommended as the fastest, cheapest, most flexible and effective way for identifying
usability problems (Saul Greenberg et al., 2000). Also, heuristic evaluation is especially valuable when time
and resources are short (Laurie Kantner and Stephanie Rosenbaum, 1997). Jeffries and his colleagues
compared four different techniques used to evaluate a user interface. These were heuristic evaluation,
software guidelines, cognitive walkthroughs and user testing. Overall, the heuristic evaluation technique
identified the most usability problems (Jeffries, 1991). In addition, heuristic evaluation has a number of other
more specific advantages, such as low cost relative to other evaluation methods, advanced planning not
required, can be used early in the development process, may be used throughout the development process,
effective identification of major and minor problems, during a short session, a small number of experts can
identify a range of usability problems, and because of their experience with many system interfaces, it is easy
for evaluators to suggest solutions to the usability problems (Nielsen, 1990) (Holzinger, 2005) (Jenny Preece
et al., 1993).
In contrast, there are a number of major drawbacks to heuristic evaluation, such as: it is highly dependent
on the skills and experience of the evaluators, the evaluators only emulate users (they are not actually users
themselves), user feedback can only be obtained from laboratory testing, and it does not essentially evaluate
the whole design because there is no mechanism to ensure the whole design is explored (evaluators can focus
too much on one section). Also, it does not include end users, and thus is unable to specifically identify users
needs. Furthermore, it produces a large number of false positives that are not usability problems, which
IADIS International Conference WWW/Internet 2010
requires evaluators to spent extra time in elimination, and it has also been criticised as being too abstract
(Laurie Kantner and Stephanie Rosenbaum, 1997) (Holzinger, 2005) (Jacob Nielsen, 1992) (Jeffries, 1991).
Therefore, there have been a number of attempts to avoid these heuristic evaluation weaknesses, and these
will be discussed in the following section.
2.3 The Need for a Customised Guidelines Method
The need for customised guidelines appeared when a number of studies, such as Chen and Gavriel or Chen
and Macredie, extended the traditional heuristic evaluation method, in different ways, to match their
evaluation needs for different computing systems (as some systems may have more priorities for some
usability criteria than other systems), and to avoid the drawbacks of the heuristic evaluation method. The
modifications occurred in one of three ways: extending the heuristic set, extending the heuristic evaluation
method by modifying the evaluation procedure, and extending the heuristic evaluation method with a
conformance rating scale (Chen and Gavriel, 2005) (Chen and Macredie, 2005).
2.4 Proposing the Customised Guidelines
The customised guidelines method is a combination of two methods; heuristic evaluation method plus plus
(HE++) and heuristic evaluation plus (HE+). HE++ is one of the proven methods of extending the heuristic
set. A number of studies have shown that HE++, an emerging variant of the original HE, outperforms HE in
both effectiveness and reliability (Jarinee and Gitte, 2008) (Jarinee and Jaqueline, 2004). HE+ includes an
extra feature for aiding the evaluation process, called a usability problem profile (Jarinee and Jaqueline,
2004) (Jarinee and Gitte, 2008), which is designed to address the fact that many of the problems found by
evaluators are common to certain types of products or applications. The proposed method targets commercial
websites, although it will employ the same technique of heuristic evaluation in terms of using guidelines and
evaluators only, and does not involve any extra resources. Thus, it can be now stated that the customised
guidelines method consists of combination of the HE++ and HE+; the first component is the HE++, which
has the following criteria derived from Jarinee and Gitte (2008):
1. Visibility, distinguishability, and timeliness
2. Match between system and the real world
3. User control and freedom
4. Consistency and standards
5. Error prevention and recovery
6. Recognition rather than recall
7. Flexibility and efficiency of use
8. Aesthetic and minimalist design
9. Supports user tasks and avoids difficult concepts
10. Supports modification and progress evaluation
11. Relevant, correct, and adequate information (Jarinee and Gitte, 2008).
The second component is the HE+ method, which includes a usability problem profile that can be
fruitfully used in many instances. Moreover, several researchers have discussed the components that should
be included in the usability problem profile for e-commerce websites. Chattratichart and Lindgaard reported
that the most recent profile now consists of seven problem areas: content, navigation, graphics, system
efficiency and functionality, formatting and layout, wording, and help and error messages (Jarinee and Gitte,
2008). However, Chattratichart and Brodie state that e-commerce websites should include only six problem
areas: content, graphics, navigation, layout, terminology, and matches with users tasks (Chattratichart and
Brodie, 2002) (Chattratichart and Brodie, 2004). In addition, Van der Merwe and Bekker said that when an
evaluator evaluates an e-commerce website, the criteria or areas should include: content, graphics,
navigation, reliability, and security and privacy (Rian Van der Merwe and James Bekker, 2003). After
studying all of these, the researchers here decided to choose the components for the usability problem profile
(in e-commerce websites) based on those criteria that are present in the majority of these studies. The final
usability problem profile in this research contains Content, Graphics, Navigation, Layout and Formatting,
Reliability, and Privacy and Security.
ISBN: 978-972-8939-25-0 2010 IADIS
The primary aim of this research is to compare specific customised guidelines with the more traditional
heuristic method; the Nielsen heuristic set is chosen to establish which method can offer better results. In
order to achieve this aim, there are some objectives that need to be achieved as well. The following is a
description of the research objectives.
1. To propose a customised guidelines method for measuring the usability of commercial websites.
2. To apply the customised guidelines and the Nielsen heuristic set to targeted websites.
3. To compare the customised guidelines vs. heuristic evaluation for commercial websites in order to
reveal each methods performance.
Two main experiments were undertaken in order to achieve the research objectives. The first experiment
was to apply heuristic evaluation on two commercial website by using two different groups of evaluators;
each group performed HE on a website. The second experiment was to apply the new customised guidelines
method upon the same two commercial websites by the same two groups. The group G1 performed HE first
on Website A, and then performed the customised guidelines on Website B. The group G2 performed the
customised guidelines on Website A, and then performed HE on Website B. In more detail, each group
started with a different website in order to further control the experiment, rather than performing one method
at the beginning of both studies. Each evaluator had a questionnaire at the end of each evaluation to answer
some questions with regard to the method he had used.
3.1 Website Selection
The targeted websites should be live, dynamic, have a number of features, and be of reasonable content size.
This selection has only targeted commercial websites because of the limitation of the study. Therefore, the
researchers decided to choose two electronic shopping websites to be the targeted websites that meet the
above conditions.
3.2 Number of Evaluators
Section 2.1 discussed the number of evaluators needed to conduct a usability study. In this research, 8
evaluators were needed, where each group had 4 evaluators and they were divided evenly in terms of
experience and other factors. Therefore, each website has been evaluated by the 8 evaluators but each four
used a different method.
3.3 Severity Assessment
Usability evaluation methods depend on evaluators judgment when they rate a usability problem. Here in
this study, the rating of usability problems was based on the average rating of all the evaluators. Their
assessment is based on three factors, which are the problem impact upon business goal, its frequency and its
persistence. This should eliminate any deviation or personal judgment, and it is suggested by several studies,
such as (Chen and Macredie, 2005) and (Morten Hertzum, 2006).
This section presents the data analysis from this experiment. It discusses how well the two groups performed
while they were evaluating Website A. It then also discusses the results from the second website (B) after it
was evaluated by both groups.

IADIS International Conference WWW/Internet 2010
4.1 Website A Evaluation Results and Analysis
Group G1 used the Nielsen heuristic set whereas group G2 used the customised guidelines. Each evaluator in
G1 found a number of problems in Website A by using HE. The final list consists of 10 problems, which are
summarised in Table 1; the Total Problems by group G1 column refers to the total number of problems
identified by the four evaluators together. They were able to identify 5 major problems and the remaining
problems were minor or superficial problems. In general, group G1 was able to reveal less than a quarter of
the identified usability problems by both groups in this study. The evaluators agreed from their questionnaire
responses that the website is highly usable and has taken into account usability guidelines.
Table 1. Number and percentage of usability problems uncovered on Website A by group G1.
HE Evaluator
Total Problems
by group G1
Total problems
discovered by both
groups (G1, G2)
Total 10 2 5 3 10 42
% of
23.80% 4.76% 11.90% 7.14% 23.80% 100%

Group G2, who performed the customised guidelines method, were able to identify 40 usability problems
(summarised in Table 2), which was significantly more than the number of the problems identified by group
G1. The results from Table 2 show the percentages of the identified problems, which range from 35.71% to
61.90%. As a result, this evaluation, which is based on the customised guidelines, obtained percentages that
are in line with findings of similar studies (Jakob Nielsen, 1994, 1990), which show that a single evaluator
usually finds 20% to 50% of the problems in a system. All the evaluators in group G2 were pleased to be
using the customised guidelines method as this method reminds them of a number of usability aspects they
might have forgotten to think about during the evaluation. Also, according to the evaluators answers in the
questionnaire, they believe that the customised guidelines method encourages them to be thorough in the
evaluation, which is seen as a drawback of HE, providing more opportunities for finding usability issues. The
evaluators answers in group G1 state that HE did not help much as it is so general, does not cover all the
usability aspects, and does not offer any clues for the evaluation. Although, the website was claimed to be a
highly usable, group G2 was able to uncover different usability problems, including 8 major problems and 25
minor problems.
Table 2. Number and percentage of usability problems uncovered on Website A by group G2.
Total Problems
by group G2
Total problems
discovered by both
groups (G1, G2)
Total 21 26 15 19 40 42
% of Total
50% 61.90% 35.71% 45.23% 95.23% 100%
4.2 Website B Evaluation Results and Analysis
Group G1, who used the Nielsen heuristics in the first study, used the customised guidelines method in this
study to offer more validity to the study. Each evaluator in group G1 found a number of problems in Website
B. The final list of the identified usability problems consists of 38 problems, which are summarised in Table
3. The results from Table 3 show the percentages of the identified problems, which range from 39.47% to
74.47%. As a result, this evaluation, which is based on the customised guidelines, obtained percentages
which are in line with and even higher than the findings of similar studies (Nielsen, 1990), which show that a
single evaluator usually finds 20% to 50% of the problems in a system. They discovered 5 major usability
problems and 20 minor usability problems. The evaluators attributed this to the usage of the customised
guidelines method as they preferred to use it rather than to use the HE method. They justified this as the
former method clearly specifies what should be evaluated and it is also a useful technique of building a
ISBN: 978-972-8939-25-0 2010 IADIS
checklist; they also liked the idea of the usability problem profile as it seems to work as a double-check for
the identified usability problems.
Table 3. Number and percentage of usability problems uncovered on Website B by group G1.

Group G2, who used the customised guidelines in the first study, then used the HE method to evaluate
Website B. Each evaluator in this group was able to reveal a number of problems in Website B. The results
are summarised in Table4. They were able to reveal more than 21% of the total of the identified usability
problems. The results ranged from 10% to 21% of the total number of the problems. Comparing these results
to group G1 using the same tool in the previous results, there is not a huge difference as both of them were
not able to identify more than 23% of total number of the identified usability problems.
Table 4. Number and percentage of usability problems uncovered on Website B by group G2.

In addition, 7 out of the 8 evaluators stated that they would prefer to use the customised guidelines
method in future evaluations rather than the HE method. The given reasons for this were that the customised
guidelines method is easier to understand and use than HE. They also stated that the customised method
seems to encourage them to be more through. On the other hand, evaluators spent less time when they used
HE for both websites, they spent 40 and 37.5 minutes for A and B websites respectively. The time spent on A
and B websites by the evaluators while they used the customised method were 72.5 and 50 minutes
respectively. Therefore HE seems to be faster than the customised guidelines method. Both of the methods
performed well in discovering major problems. HE has been criticised as being an abstract method, which
does not provide guidelines for a specific domain, for example education or government websites.
Generally, it can be seen that specific guidelines are preferred to the general guidelines as they can help
evaluators to focus on those criteria that are important to a particular type of a website in order to integrate all
the most suitable usability considerations. For instance, in this research, the customised guidelines for
electronic shopping emphasise graphics and security, which may be less important in, for example,
educational websites. Moreover, using a checklist in the guidelines maximises the number of usability
problems that can be identified for each interface. Furthermore, the use of specific guidelines can break down
the usability problems identified to the lowest level of detail. Also, specific guidelines can provide a more
comprehensive analysis of the interface usability problems.
It can be seen clearly from the literature and from the findings of this paper that there is a clear need to have
customised guidelines for specific domains such as commercial, education or government websites. In this
paper, the customised guidelines for commercial websites performed better and were preferred by most of the
evaluators. There are a number of advantages to the customised guidelines method: it is a comprehensive and
detailed method at the same time, and it seems to work as a checklist for the usability guidelines. However, it
has a number of drawbacks: it is limited to commercial websites, and it seems to lead the evaluators to the
website aspects rather than leaving them to look over the site in general and think out of the box.
Total Problems
by group G1
Total problems discovered
by both groups (G1, G2)
Total 28 22 15 20 34 38
% of Total
74.47% 57.89% 39.47% 52.63% 89.47% 100%
HE Evaluator
Total Problems
by group G2
Total problems discovered
by both groups (G1, G2)
Total 5 8 4 5 12 38
% of Total
13.15% 21.05% 10.52% 13.15% 31.57% 100%
IADIS International Conference WWW/Internet 2010
Chen, L. & Gavriel, S. (2005). Extension of heuristic evaluation method: a review and reappraisal. Ergonomia IJE\&HF
27: pp.179-197.
Chen, S. & Macredie, R. (2005). The assessment of usability of electronic shopping: A heuristic evaluation. International
journal of Information Management 25: pp.516-532.
Hix, D., Swan, JL., Hollerer, M., Julier, T., Baillot, S., Brown, Y., Center, D. and Blacksburg, V. (2004) A cost-effective
usability evaluation progression for novel interactive systems. the 37th Annual Hawaii International Conference on
System Sciences. IEEE Computer Society, Washington, USA.
Holzinger, A. (2005). Usability engineering methods for software developers. Communications of the ACM 48: pp.7174.
Jakob Nielsen & Rolf Molich (1990) Heuristic evaluation of user interfaces. Proceedings of the SIGCHI conference on
Human factors in computing systems: Empowering people. ACM, Seattle, Washington, United States.
Jakob Nielsen (1994) Guerrilla HCI: using discount usability engineering to penetrate the intimidation barrier. Cost-
justifying usability. Orlando, FL, USA. pp. 245--272.
Jakob Nielsen (2005) Heuristic Evaluation. In:
Jacobsen Nielsen (1992) Finding usability problems through heuristic evaluation. the SIGCHI conference on Human
factors in computing systems. ACM, NY, USA, pp. 373380.
Jarinee, C. & Gitte, L. (2008) A comparative evaluation of heuristic-based usability inspection methods. CHI '08
extended abstracts on Human factors in computing systems. ACM, Florence, Italy.
Jarinee, C. & Jaqueline, B. (2002) Extending the heuristic evaluation method through contextualisation. Human Factors
and Ergonomics Society 46 th Annual Meeting, pp. 641-645.
Jeffries, R., Miller, J., Wharton, C., and Uyeda, K. (1991) User interface evaluation in the real world: a comparison of
four techniques. The SIGCHI conference on Human factors in computing systems. ACM, NY, USA, pp. 119-124.
Jenny Preece, Gordon Davies, David Benyon, Laurie Keller & David Benyon (1993). A guide to usability: Human
factors in computing MA, USA: Addison-Wesley Longman Publishing Co.
Koyani, S., Bailey, R., Nall, J., Allison, S., Mulligan, C., Bailey, K., Tolson, M., (2003) Research-based web design &
usability guidelines.
Laurie Kantner & Stephanie Rosenbaum (1997) Usability studies of WWW sites: Heuristic evaluation vs. laboratory
testing. The 15th annual international conference on Computer documentation, ACM NY, USA, pp. 153-160.
Morten Hertzum (2006). Problem Prioritization in usability Evalution: From severity assessments toward impact on
design. International Journal of Human-Computer Interaction 21: pp.125-146.
Morten Hertzum & Nielsen Ebbe Jacobsen (2001). The Evaluator Effect: A Chilling Fact About Usability Evaluation
Methods. International Journal of Human-Computer Interaction 15: pp.183-204.
Rian van der Merwe & James Bekker (2003). A framework and methodology for evaluating e-commerce web sites,
Internet Research: Electronic Networking Applications and Policy 13: pp.330-341.
Saul Greenberg, Geraldine Fitzpatrick, Carl Gutwin & Simon Kaplan (2000). Adapting the locales framework for
heuristic evaluation of groupware. Australian Journal of Information Systems 7: pp.102108.
Scott Henninger (2001) An Organizational Learning Method for Applying Usability Guidelines and Patterns.
Proceedings of the 8th IFIP International Conference on Engineering for Human-Computer Interaction. Springer-
ISBN: 978-972-8939-25-0 2010 IADIS
Gerald Hbsch, Christian Liebing, Josef Spillner and Alexander Schill
Technische Universitt Dresden, Institute for Systems Architecture, Chair of Computer Networks
In this paper, we present the User Interface Service Description Language (UISDL) for the description of UI Services. UI
Services allow UI component implementations to be globally distributed, flexibly integrated and dynamically bound into
interactive applications including, but not limited to, Rich Internet Applications. We present abstraction concepts,
motivate design decisions, and discuss analogies between UISDL and WSDL as a classical service description language.
UI component, UISDL, user interface services, WSDL, XML Schema
While the service-based provisioning of business logic and data has reached a high degree of maturity and
acceptance in the industry, the advantages of service-orientation for other building blocks of Rich Internet
applications (RIAs), especially their user interface (UI), have not yet been sufficiently addressed. Application
developers are today confronted with the need to choose one out of a large number of UI framework
providers, which in many cases may be already compromise. Due to a lack of standardization, they then face
a 'vendor lock-in' which makes it difficult to flexibly integrate or exchange complex user interface
components like maps, tables, calendars, or chart viewers, from third parties. Equally, providers of UI
components face difficulties in providing them 'as-a-service' i.e. in a standardized self-descriptive way that
enables their flexible integration, distribution and retrieval. We therefore believe that novel concepts are
necessary that make UI components available as UI Services. These concepts must address challenges that
relate to the self-description of UI components that abstract from vendor- and implementation-specific
details, mechanisms for their flexible integration into RIAs and their distribution on a global scale.
The CRUISe project (Composition of Rich User Interface Services) addresses the challenges named
above. The key idea is to enable UI components to be described and distributed as UI Services that can be
flexibly composed into user interfaces for RIAs. The main benefit of UI Services is that UIs of RIAs can
dynamically bind the most suitable UI component implementation at runtime based on context, e.g. the
platform type or user preferences. In this paper, we present the User Interface Service Description Language
(UISDL) as a central technological result of the project. UISDL is an XML format for describing UI
components as a set of abstract events, operations, and properties that define UI component interfaces in an
implementation- and platform-independent way. UISDL supports the binding of these abstract interface
definitions to alternative component implementations provided by different vendors and for multiple
platforms, e.g. special versions for desktops and mobile phones. UISDL includes metadata for the
management and distribution of UISDL descriptions through a UI Service repository. This paper is structured
as follows. Section 2 discusses related work. Section 3 presents UISDL together with the related concepts.
Our runtime architecture for UI Services based on UISDL and sample applications built on top of UI
Services are presented in section 4. We conclude our work and outline future work in section 5. To facilitate
understanding, we will use an electronic map as an exemplary UI component throughout this paper.

IADIS International Conference WWW/Internet 2010
The analysis of related approaches shows that little research has been done to facilitate service-based
provision and dynamic integration of UI components using description languages like UISDL. The Mixup
approach (Yu et al. 2007) introduces a development and runtime environment to compose web applications at
the presentation layer based on existing UI components. The abstract UI-only component model enables
integration into existing composite applications and defines the presentation state, properties to represent the
appearance, events to signal state changes and operations to react to state changes triggered by other UI
components. To describe UI components, the approach uses XPIL (Extensible Presentation Integration
Language) as a declarative description language that solely contains a small set of elements describing the
abstract component model similar to WSDL. Unlike UISDL, XPIL solely describes the interface of UI
components and does not distinguish between an abstract and concrete part to enable the distribution and
flexible integration of different component implementations. Furthermore, component metadata is not
described, making the registration and discovery in a repository complicated. We will show that especially
the consideration of these two aspects is vital for the realization of UI Services.
Since the Mixup approach only provides a UI-only component model, the authors introduced a platform
for universal composition called mashArt (Daniel et al. 2009). The universal component model extends the
model presented in (Yu et al. 2007) and encapsulates UI, application and data components by defining their
state, properties, events and operations. The specification of a component has the form of an abstract
component descriptor as a main part of the mashArt Description Language (MDL), which has a very similar
structure to XPIL, but aims at a universal component description. However, MDL does not provide metadata
to enable its registration and discovery in a repository. Furthermore, it lacks a separation into abstract and
concrete parts, leading to the above-mentioned drawback.
The Service-Oriented User Interface Modelling and Composition (SOAUI) approach (Tsai et al. 2008)
proposes the discovery and dynamic integration of UI components into a system design process of service-
oriented applications at runtime. The specification is based on proprietary Microsoft XAML (Microsoft
2010) and describes the offered functionality of a UI component. The authors intend to establish a unified UI
specification standard to share components between heterogeneous systems and platforms. They mainly
focus on the management and retrieval of UI components, while our work explicitly includes their service-
based provision and dynamic integration. Widgets (Cceres 2009) or gadgets (Google 2008) are artifacts that
statically combine a piece of application logic with a UI component to a small application. They could
benefit from UI Services described with UISDL, since the choice of the concrete UI component
implementation to be used for the widget can be left open until runtime, leaving it up to personal preferences
of the user or the technical context in which the widget/gadget is used.
In summary, only very few approaches exist that use a declarative description language for UI
components. Unlike UISDL, these approaches do not sufficiently address the service-based provision and
dynamic integration of UI components through UI Service regarding the separation of concrete and abstract
parts, explicit support for metadata, and for dynamic integration at runtime.
This section presents the UISDL in detail. Starting with the introduction of our UI component model and a
discussion of requirements for the language, design decisions and language concepts are presented. We also
mention language concepts that are inherited from WSDL as a classical service description language.
3.1 UI Component Model
The basis for UI Services, and thereby for UISDL, is the specification of a generic UI component model. Our
UI component model abstracts from implementation details by defining UI components as a set of abstract
properties, events, and operations.
Properties represent the externally visible UI component state. Common properties of an electronic map
are, for example, the geographic coordinate of its center or its type (normal, satellite, hybrid). Events are
triggered by UI components to publish state changes to other components, usually in response to user
ISBN: 978-972-8939-25-0 2010 IADIS
interaction. In the electronic map, a centerChange event is triggered in response to the respective action of
the user. The functionality of a UI component is represented by operations that consume input parameters
and produce at most one return parameter. A sample operation of the electronic map is addSimpleMarker. It
adds a Point-of-Interest marker at the geo-coordinate supplied as an input parameter.
3.2 Requirements
The purpose of UISDL is to define an XML-based description language for UI Services. UISDL therefore
has to provide means to describe the interfaces of UI components that follow the UI component model
presented in section 3.1, including the data types they use to communicate with the application that integrates
them in an abstract, i.e. implementation- and programming language independent, way.
To distribute component implementations, UISDL must support the binding of component
implementations to an abstract interface description. Implementation bindings must (a) enable the life cycle
management of UI components, (b) provide access to properties, (c) enable the invocation of component
operations, and (d) allow the (de-)registration of event listeners. Furthermore, libraries that must be included
to use the component implementation must be specified as part of the implementation binding.
To fully exploit the flexibility offered by UI Services, it must be possible to implement runtime
mechanisms on top of the implementation bindings that support a dynamic binding of UI Services. Dynamic
binding enables the selection and exchange of a concrete UI service implementation at runtime, for example
based on the context of use, including user preferences and platform properties. For the same reason, abstract
interface definitions and implementation bindings must be separated since multiple UI component
implementations, e.g. Google Map, Yahoo Map, OpenStreetMap, offered by different providers can exist for
one abstract Map interface. It is clear that their common abstract interface can only cover the common set of
functionalities offered by the implementations. However, we have found that existing component
implementations often extend this common set with additional functionalities. For example, some existing
implementations of the abstract Map interface, like Google Maps, allow to drag-and-drop of Point-of-Interest
markers and produce a corresponding event in response. This requires the specification of an abstract
interface specifically for these implementations that extends the Map interface with the additional event.
Otherwise, the additional functionality would either not be accessible or applications using the Map UI
Service could not detect the compatibility of the extended interface with Map UI. UISDL must therefore
support the declaration of functional extensions between abstract interfaces.
Finally, we have found inline metadata in the form of keywords, screenshots of UI component
implementations, license information, pricing, suitability, and human-readable documentation to be necessary
for the management of UISDL descriptions in a UI Service repository (see sect. 4) and their integration in
composition tools.
3.3 Language Concepts and Design Decisions
The language concepts and design decisions for UISDL are guided by the requirements in sect. 3.2.
Interestingly, the requirements that have to be fulfilled by UISDL regarding abstraction, binding and
integration into applications are quite similar to those addressed by WSDL. The major difference seems to be
that web services are executed remotely on a server, while UI components provided by UI Services are
executed locally in the scope of the application that integrates them. For UI Services, the need to provide an
execution environment on the client that manages the UI components' life-cycle replaces the need for remote
service invocation mechanisms required for web service integration. As we will show in the following,
concepts developed for classical web services that have shown to be reusable for our approach.
WSDL is structured into an abstract part and a concrete part. Its abstract part specifies the web service
interface, i.e. the operations offered by the service, and the data type definitions they use in a platform-,
protocol- and programming language independent way against which applications can be developed. Its
concrete part contains protocol binding information and endpoint definitions that allow applications to invoke
a specific implementation of the web service. UISDL adopts these abstraction and structuring concepts by
defining two different document types, namely UISDL-Class (UISDL-C) for the abstract part and UISDL-
Binding (UISDL-B) for the concrete part. By defining two different document types for both parts, we go one
step beyond WSDL. Their physical separation was chosen to provide the organizational means for a central
IADIS International Conference WWW/Internet 2010
management and distribution of abstract UI component interface definitions that can be referenced by
providers of compatible UI component implementations. It thereby allows providers of UI component
implementations as well as developers that integrate UI Services to ensure compatibility on a global scale.
UISDL-C is used to define abstract UI component interfaces: properties, operations, events, and data
types in an implementation- and programming language independent way similar to WSDL. It fully reflects
our UI component model. Syntax details are presented in sect. 3.4. Applications that use UI Services are
programmed against the abstract interfaces defined with UISDL-C.
UISDL-C also adopts the data type definition based on XML Schema as an abstraction concept from
WSDL. XML Schema was chosen due to its programming language independence. Like in WSDL, XML
Schema definitions are utilized to specify the data types required for the abstract interface definition, i.e.
properties, invocation and result parameters of operations, and events. Furthermore, instances of XML
Schema data types can easily be serialized in XML. We have therefore decided to use XML as a common
and implementation-independent data exchange format between UI component implementations to maintain
implementation independence at runtime. This is the prerequisite that UI component implementations must
fulfill to be distributable as UI Services. It is the responsibility of each component implementation to convert
between the XML representation and internal data representations. Since the data exchange with web
services via SOAP uses the same strategy, our approach also simplifies the integration between UI Services
and web services, since UI component data can be directly wrapped in SOAP messages.
With UISDL-B, implementation bindings for concrete UI component implementations are specified and
distributed. They are created by UI component providers and bind the abstract interface defined in UISDL-C
to implementation-specific component operations for life cycle management (constructor, destructor), for
showing and hiding the UI component, accessors for properties, component operations, and for
(un)registering event handlers. Analogous to WSDL, references from UISDL-B to constituents of the abstract
interface being bound are based on names. Also, UISDL-B includes elements to specify the libraries to be
included before the component can be instantiated. Syntax details and realizations for bindings are presented
in sect. 3.5. The abstract interface implemented by a UISDL-B description is denoted by a reference the
corresponding UISDL-C. Inverse references from UISDL-C descriptions to UISDL-B implementation
bindings are avoided since the number of component providers that can vary over time, which imposes the
risk of invalid references and would require frequent updates of UISDL-C descriptions. Instead, the UI
Service repository (sect. 4) stores and updates these references.
The possibility to dynamically select, bind, and even exchange component implementations delivered by
UI Services does not require them to be compatible on the technical interface level, which is ensured by the
abstract interface definitions. However, interface compatibility alone does mean that all implementations
provide the same functionality. Therefore, also the functional equivalence of different component
implementations for one abstract interface must additionally be considered. Since this is hardly possible to
guarantee from a technical point of view, UISDL-C supports an organizational model based on a functional
classification which allows component providers to express functional equivalence. A detailed discussion of
our functional classification approach and their definition goes beyond the focus of this paper. Briefly, we
define one class per abstract interface definition, i.e. UISDL-C document that specifies the functionality to be
provided by its implementations. UISDL-C supports this classification through a mandatory reference to its
class. For the same reason, the name of the abstract part of UISDL was chosen to be UISDL-Class.
As motivated in the requirements, it is also necessary to provide means to declare functional extension
relationships between two abstract interfaces as part of UISDL-C descriptions. In the drag-and-drop
extension example, it is obvious that the Google Map implementation (which supports drag-and-drop), is
functionally equivalent to the Yahoo Map (which does not feature drag-and-drop support) in applications
implemented against the abstract Map UI interface. The presence of a functional extension is based on two
criteria: interface compatibility and functional equivalence of the extending interface's class to the class of
the extended interface. Informally, by declaring an abstract component interface B to be a functional
extension of an abstract UI component interface A, it is expressed that B adds at least one element to the set
of operations, properties, events, or data types defined in A, without redefining other elements in I
. I.e. B
technically remains interface compatible with A. The second criterion of functional extension requires
implementations of B to implement equivalent functionality for all elements in I
, i.e. to adhere to A's class,
its superclass. Through this concept, we can specify that any implementation of B can be used as an
alternative for an implementation of A, thereby increasing the number of alternatives for the dynamic
binding. UISDL-C therefore features an optional superclass reference.
ISBN: 978-972-8939-25-0 2010 IADIS
Metadata for management, retrieval and documentation purposes can be added to UISDL-C and UISDL-
B descriptions. Both support a keyword list for keyword-based indexing in the UI Service repository and
global documentation. UISDL-C descriptions additionally allow documentation of each constituent of the
abstract interface. UISDL-B supports hyperlinks to screenshots of the UI component, e.g. for its presentation
by a UI Service repository browser, a list of devices that are supported by the component implementation
including a suitability value to simplify context-aware UI Service selection. Furthermore, basic service
pricing information and a license text can be added for commercial purposes.
3.4 UISDL-class
The syntax of UISDL-C is presented by the Map abstract interface description example in listing 1. For
reasons of brevity, only one example per element is shown. The attribute class of the root element denotes the
functional classification of the abstract interface description. Furthermore, global documentation (line 2),
metadata (lines 3-5), XML Schema data type definitions (lines 6-17), and the interface (lines 18-30)
specifying the properties (lines 19-22), events (lines 23-25), and operations (lines 27-29) are defined in the
UISDL-C. Properties, events, operations, and parameters are identified by unique names (name attributes).
Properties and parameters are typed by XML Schema data type. When used as child elements of operation,
parameter elements specify the operation's invocation and result parameters. As child elements of event, they
specify data belonging to the event, e.g. the map center's geographic coordinate after its adjustment by the
user (centerChange event).
3.5 UISDL-binding
The syntax of UISDL-B is presented by a UI component implementation that uses the Google Map API to
implement the Map abstract interface description presented in sect. 3.4 in listing 2. The document is
composed of global documentation (line 2), a reference to the implemented class (line 3, cp. sect. 3.4),
metadata (lines 4-18): component screenshots, device classes the implementation supports, including s
1 <uisdlc class="/geo/map" xmlns="">
2 <documentation>A UI Service class for map components</documentation>
3 <metadata>
4 <keywords>map, maps, geo, orientation, directions, POI</keywords>
5 </metadata>
6 <datatypes>
7 <xs:schema xmlns:xs=""
8 xmlns:def="" targetNamespace="">
9 ...
10 <xs:complexType name="GeoCoordinate">
11 <xs:sequence>
12 <xs:element minOccurs="1" maxOccurs="1" name="lat" type="def:latitude.type"/>
13 <xs:element minOccurs="1" maxOccurs="1" name="long" type="def:longitude.type"/>
14 </xs:sequence>
15 </xs:complexType>
16 </xs:schema>
17 </datatypes>
18 <interface xmlns:def="">
19 <property type="def:GeoCoordinate" name="">
20 <documentation>The GeoCoordinates of the center of the map.</documentation>
21 </property>
22 ...
23 <event name="event.centerChange">
24 <parameter type="def:GeoCoordinate" name="center"/>
25 </event>
26 ...
27 <operation name="op.addSimpleMarker">
28 <parameter type="def:GeoCoordinate" direction="in" name="markerPos"/>
29 </operation>
30 </interface>
31 </uisdlc>
Listing 1. UISDL-class example
IADIS International Conference WWW/Internet 2010
suitability attribute to support the selection process for dynamic binding at runtime, keywords, license
information, and pricing. The interface binding element (lines 19-49) specifies information necessary to
integrate, access, and manage the component life cycle at runtime for the specific implementation being
bound. The language attribute holds the programming language of the component implementation. The
binding type attribute denotes the type of implementation binding used.
The dependencies element (lines 20-23) specifies the libraries that must be included to use the
component. The elements constructor and destructor are utilized for component life-cycle management, the
rendering element specfies how the component is set visible. An accessor element exists for each component
property specified in the UISDL-C and specifies read and write access. For each event element in UISDL-C,
the eventsink element specifies how event listeners are registered and unregistered. For each operation in the
UISDL-C, the invocation element specifies how the operation is invoked. The attributes property, event, and
operation are references from UISDL-B to UISDL-C. They are generally based on the interface element
names specified in the corresponding UISDL-C. Therefore, applications can be implemented against the
abstract interface specification.
By now, we have identified the two binding types code template and convention as being suitable for
UISDL-B. The example in listing 2 uses code templates, which are applicable for script languages like
Javascript that support execution of code generated at runtime. Briefly, in the code templates, placeholders
enclosed by '@' symbols represent application-specific parts that must be replaced with application-specific
values (constants, variables, pointers) before the code can be executed. Placeholder types to refer to events,
properties, and parameters specified in the UISDL-C (e.g. @paramater:center@), or to the component
instance (@:instance@) have been defined. This enables an application to pass these values to a code
template processor using the names in the UISDL-C. We have found these placeholder types to be sufficient.
Code templates are enclosed in code elements (cp. fig. 2). The convention binding type directly maps the
abstract interface specified in UISDL-C to a concrete implementation skeleton, e.g. through XSLT. The
component implementation can in this case be managed and accessed with knowledge about the mapping
rules. The mapping rules used must therefore be identifiable from the interface binding's binding type
A client-side thin server runtime (TSR) that provides the runtime mechanisms required to support integration
and dynamic binding of UI Services has been implemented for web browsers (Pietschmann et al. 2010). It
provides RIAs with the necessary runtime mechanisms to dynamically bind UI component implementations
described in UISDL-B, to manage their life-cycle, and to access them from client-side business logic. The
TSR supports the binding types code templates and convention. Furthermore, we realized a repository to
register, manage and retrieve UI components and to enable their global distribution.
For UI component implementations described in UISDL-B, the TSR provides means to create, render,
and destroy UI component instances, to access their properties, to add and remove event handlers, and to
invoke their operations in a uniform way. The selection of the UI Service to be bound is either done by
passing a UISDL-C class name or the URL of a UISDL-B to the TSR. It queries the repository to retrieve a
UISDL-B document using AJAX. If a class name is used, the selection of an appropriate UISDL-B is
delegated to the repository and performed based on information about requesting device that is passed with
the query, including an option to preselect a concrete component implementation. Upon receiving the
UISDL-B, the TSR creates the UI component instance.
Application-specific information required by the Integration Manager, i.e. variables, constants, and
pointers to event handler functions, are passed from the business logic via API calls programmed against the
abstract UISDL-C interface specification, e.g. the name of the property to be set together with the new value,
and an identifier by which the component instance is registered by the Integration Manager.
To conveniently integrate and initialize UI component implementations, we have defined UI component
placeholders that allow service selection information together with application-specific information be
specified as part of HTML documents which we term composition documents. The TSR is able to process
composition documents, making the integration of UI Services without programming particularly easy.
Details of the composition document are beyond the scope of this paper.
ISBN: 978-972-8939-25-0 2010 IADIS
The UIS repository is implemented as a Web Service that offers operations to register, search, and
distribute UISDL-C and UISDL-B descriptions. UISDL-C descriptions are registered based on their meta
information and classification, including the class/subclass relationship (cp. sect. 3). The repository also
stores UISDL-B descriptions. In the repository, they are associated with registered UISDL-C descriptions
based on the class references they contain. When queried by the Integration Manager using a UISDL-C class
name and device information, the repository is able to the select and return an appropriate UISDL-B
document in response, or an error if no appropriate implementation binding was found.
Another possibility is to use the repository to find UISDL-C or UISDL-B descriptions during the
implementation of RIAs that use UI Services. In this case, the possibility for a keyword-based search and the
integration of screenshots in UISDL-B have shown to be useful for browsing the repository.

To validate our approach, two RIAs have been successfully implemented based on UISDL and the TSR: a
travel planner application for public transport that retrieves timetable information from a web service back
end and a real estate management application. Their UIs are built using UI Services that deliver tables, maps,
1 <uisdlb xmlns="">
2 <documentation>Google Map UI Service implementation</documentation>
3 <uisdlclasses>/geo/map</uisdlclasses>
4 <metadata>
5 <screenshots>
6 <screenshot>
7 <documentation>Google Map Component Preview</documentation>
8 <uri></uri>
9 </screenshot>
10 </screenshots>
11 <devices>
12 <deviceclass name="pda" suitability="0.6"/>
13 ...
14 </devices>
15 <keywords>map, maps, geo, orientation, directions, Google</keywords>
16 <license>Text of the license agreement.</license>
17 <pricing unit="USD" price="0.00"/>
18 </metadata>
19 <interfacebinding language="javascript" bindingtype="codetemplates">
20 <dependencies>
21 <dependency uri="" language="javascript"/>
22 <dependency uri="" language="javascript"/>
23 </dependencies>
24 <constructor><code>
25 @:instance@ = new GMap({width: @property:property.width@, height:
26 @property:property.height@, title: @property:property.title@, center:
27, zoomLevel: @property:property.zoomLevel@,
28 mapType:@property:property.mapType@})</code>
29 </constructor>
30 <destructor><code>@:instance@.destruct()</code></destructor>
31 <rendering>
32 <code>@:instance@.render(@property:property.rendertargetid@)</code>
33 </rendering>
34 <accessor property="property.width" type="get">
35 <code>@:instance@.getWidth()</code>
36 </accessor>
37 ...
38 <eventsink event="event.centerChange">
39 <register><code>
40 @:instance@.on('centerChange', @event:event.centerChange@)
41 </code></register>
42 <unregister><code>
43 @:instance@.un('centerChange', @event:event.centerChange@)
44 </code></unregister>
45 </eventsink>
46 <invocation operation="op.addSimpleMarker"><code>
47 @:instance@.addCustomMarker(@parameter:gLatLong@, @parameter:image@,
48 @parameter:imageSize@, @parameter:shadow@)</code></invocation>
49 </interfacebinding>
50 </uisdlb>
Listing 2. UISDL-binding example
IADIS International Conference WWW/Internet 2010
chart viewers, and timetable query input UI components. Their application logic is implemented in
JavaScript. Code templates and convention-based bindings are utilized in the UISDL-B documents of the UI
Services. The implementation of these applications has shown that our concepts and design decisions for
UISDL are sound and applicable for the description and distribution of the complex UI components used in
the prototypes. Also, we were able to verify that complex user interfaces for RIAs can be built based on UI
Services and that UI Services can be easily integrated with client-side logic implemented in JavaScript as
well as web service based on the mechanisms provided by the TSR. Open issues are discussed in the next
In this paper, we presented UISDL as a building block for UI Services. Furthermore, a runtime environment
has been implemented that supports the proposed approach. The presented abstraction concepts and design
decisions, specifically the physical separation of abstract and concrete parts, have shown to be sound, suitable
and complete for the description of UI Services. The binding concepts based on code templates and
conventions have shown to work in practice. The data exchange between UI component implementations
based on XML Schema instances allows for a maximum of programming language independence, but causes
processing overhead and additional effort for data adapters, especially when existing UI components are
made available as UI service implementations. JSON may be an alternative as soon as schemata for JSON
have reached a level of maturity.
Performance tests, semantic annotations, and details of the context-dependent selection of UI Services
will be subject to future work. Furthermore, the generalization of UISDL towards a unified language for logic
components and context access are under discussion in the CRUISe project.
The CRUISe project is funded by the German Federal Ministry of Education and Research under
promotional reference number 01IS08034.
Daniel, F. et al., 2009. Hosted Universal Composition: Models, Languages and Infrastructure in mashArt. In Laender,
A.H.F., Castano, S., Dayal, U., Casati, F. & Oliverira, J.P.M. (eds), Conceptual Modeling - ER. Volume 5829/2009 of
LNCS. Berlin: Springer, pp. 428-443.
Tsai, W.T. et al., 2008, Service-oriented user interface modeling and composition. Proceedings of the IEEE International
Conference on e-Business Engineering (ICBE '08). Xian, China, pp. 21-28.
Yu, J. et al., 2007, Mixup: A development and runtime environment for integration at the presentation layer.
Proceedings of the Seventh International Conference on Web Engineering (ICWE'07). Como, Italy, pp. 479-484.
Pietschmann, S. et al., 2010, A Thin-Server Runtime Platform for Composite Web Applications. Proceedings of the 5th
International Conference on Internet and Web Applications and Services (ICIW 2010). Barcelona, Spain, pp. 390-
Google, 2008, Google Gadgets Specification. Google Inc. [online] available at
Microsoft, 2010, XAML in WPF. Microsoft Corporation [online] available at
Cceres, M., 2009, Widget Packaging and Configuration. W3C [online] available at
ISBN: 978-972-8939-25-0 2010 IADIS
Lara Schibelsky Godoy Piccolo* and Maria Ceclia Calani Baranauskas**
*Fundao CPqD / IC - UNICAMP
Since Normans appropriation of Gibsons conception of affordance to explain the design of products and technologies,
this mainstream HCI concept has changed to reflect the evolution in the artifacts we interact with. This paper sheds light
on how the concept of affordance has been transformed in keeping with the changes in the HCI field. An observational
study was conducted aiming at identifying physical, perceived, social, and motivational affordances of the iPad

and of a
Tablet-PC. The study clarified how the different types of affordances can explain the relation between human and
technology in this new technical and social scenario named 3
HCI paradigm.
Physical affordances, social affordances, Human-Computer Interaction, HCI paradigms.
The field of Human Computer Interaction (HCI) has developed along its last 30 years motivated by the need
of understanding our relationship with computers and artifacts of technology in general. This relationship has
changed dramatically along that time, as a consequence of the technology design, which extrapolated the
limits of work environments, now being part of our life in almost all aspects of it. As a consequence,
mainstream concepts should be rethought, as user and technology are part of a larger system - or set of
systems. This movement of understanding design as a systemic issue has raised several debates with
implications on the fundamentals, methods and goals both for research and practice in the discipline.
As Sellen et al. (2009) suggest in redefining the H, the C, and the I in face of the transformations
the field has passed through, several levels of interaction should be taken into account: interactions on and in
the body, among people, between people and objects in the spaces of kiosks, rooms, buildings, streets and
other public areas. Central to the understanding of interaction in all these levels are the physical and social
affordances that technology can potentially enable.
Since that Norman (1988) appropriated the concept of affordance from Gibsons (1979) definition and
applied it to the design of products and technologies, this concept has been transformed and has fed a debate
about its meaning and use. This discussion originated a number of publications that explore the differences
between approaches, such as McGrenere & Ho (2000) which compared Gibsons and Normans concepts and
expanded it in a framework, and ONeill (2008) which elucidated how Gibsons affordance works, aiming at
emphasizing what he considered some authors misappropriation of the original definition.
The purpose of this paper is not to create a new comparison, but to shed light on how this term has been
transformed in keeping with the evolution in the way the human has been related to technology and the
changes in the HCI field. It is organized as follows: Section 2 presents a panorama of HCI evolution through
the three paradigms and relates it to the transformations of the concept of affordance. Section 3 discusses four
different categories of affordance considered in an observational study, described in Section 4. Discussion on
results is presented in Section 5, and, Section 6 concludes and points out future works.

IADIS International Conference WWW/Internet 2010
The HCI history has been reviewed through three main dominant paradigms built in continuous transitions,
motivated by the technology available at the time, the influence from related science fields and models and
theories used to explain those realities.
The relation between humans and computers started in the 60s, within the scenario provided by
mainframes. As the computer graphics emerged, some usability requirements were recognized as necessary
to improve the relationship of the computer with its many users (Harper et al., 2008). This technical scenario
had hardly changed in the 80s, when the Personal Computer was launched, bringing to HCI aspects of
engineering associated with human factors. All this period constituted what Harrison et al. (2007) named the
paradigm, whose main goal was to optimize the fit between humans and machines, developing
pragmatic solutions in coupling them.
In the 90s, computer networks and mobility were part of the technological domain, moving the focus to
groups working together, which constituted the focus of a 2
paradigm. Theory focused on work settings
and interaction within well-established communities of practice. Situated action, distributed cognition and
activity theory were important sources of theoretical reflection, and concepts like context came into focus of
the analysis and design of HCI. Rigid guidelines, formal methods, and systematic testing were mostly
abandoned for proactive methods such as a variety of participatory design workshops, prototyping and
contextual inquiries. (Bdker, 2006).
Many questions related to the 2
paradigm, as well as its models and theories are still current issues in
the HCI field, as Bdker (2006) argues, while a 3
paradigm started to emerge: technology spreads from
the workplace to our homes and everyday life and culture. (Harper et al., 2008).
Current phenomena related to the use of technology are transforming the society: the hyper-connectivity
by keeping people closer to the others, it may mobilize crowds in a global way; the techno-dependency in
any kind of activity; the desire to be in touch and capture information about everything; and the creative
engagement, building a society where everybody can be a content producer. People are increasingly
appropriating new digital tools, including illiterate and impaired users (Harper et. al, 2008).
Towards 2020, the technology will also be changing, according to ITEA (2009): proliferation of
embedding technology in multiple devices; sensors as input; 3D or 4D as output; augmented (AR) and virtual
reality (VR) applications; and physical machines sometimes replacing humans in interaction and decision
making. Some consequences of these changes are that new elements of human life will be included in the
human-computer interactions such as culture, emotion and experience. (Bdker, 2006) (Harrison et al.,
2007). Emotions are going to be part of context or input (ITEA, 2009); meaning is already constructed
collaboratively and interaction is influenced or perhaps even constructed by its varying physical and social
situations (Harrison et al., 2007).
In a different scientific research scenario, by the end of the 70s, the concept of affordance was created by
Gibson (1979), in his ecological theory of perception. Once it was applied to HCI, it has been transformed
following the main trends represented by the three paradigms. Nowadays, the concept of affordance has been
applied to other domains such as cognitive robotics, supporting strategies when robots are interacting with
objects (Tekkotsu, 2010). This evolution in the concept is represented in Table 1, which summarizes the main
changes in HCI field from the 70s to what has been expected by the year of 2020 in terms of technology
available, construction of meaning, predominant influence from other fields, the main question that directs
HCI researches, predominant models and theories.
Starting from the original definition of affordance, the next section describes some of the main authors
views of the concept to improve our understanding on the relationship between humans and computers.
Gibson (1979) defined the concept of affordances in the ecological context to mean [] a direct result of
the relationship between the objective physical properties of the environment and the subjective experience of
the perceiving actor within that environment. What we usually pay attention to is what an object affords us.
Based upon Gestalt theories, Gibson states that affordances are perceived with no cognitive processing, and it
is the highlight of his definition when comparing it to derived approaches.
ISBN: 978-972-8939-25-0 2010 IADIS

According to Gibson (1979), an affordance is unique for one person or animal, which makes it different
from a simple physical measure that is an object property. Affordances are objective, real and physical in a
way that "the affordance of something does not change as the need of the observer changes"; we do not need
to classify and label things in order to perceive what they afford.
To ONeill (2008), Gibson is attempting to describe an affordance as an emergent property of the
perceptual process consisting of the properties of the object itself and the perceptual capacities of the
perceiver. Different layouts afford different behaviors for different animals; a terrestrial surface, for
example, is climb-on-able, fall-off-able, get-underneath-able or bump-into-able relative to a specific animal.
Gibson also defined the term social significance to describe what one person affords to other: what
other persons afford comprises the whole realm of social significance for human beings. We pay the closest
attention to the optical and acoustic information that specifies what the other person is, invites, threatens,
and does." (Gibson, 1979).
Many authors appropriated Gibsons original term for their own uses with significant differences in the
conceptualization. The next section briefly explains some different authors perspectives on the concept.
3.1 Affordances in Design
When Norman first applied the concept of affordance in design, he stated that his conceptualization refers to
the perceived and actual properties of the thing (Norman, 1988). Those properties determine how the thing
could possibly be used. Taking advantage of it, no labels or instructions are needed. Norman (2004) proposed
a distinction between real affordances, those related to the physical properties of the world which is close
to Gibsons definition and perceived affordances which are subjective representations in the mind.
To Norman, the computer system, with its keyboard, display screen, pointing device affords pointing,
touching, looking, and clicking on every pixel of the display screen that is the real or physical affordance.
E.g.: Figure 1a: a portable computer by Sony that suggests movement and to handling with both hands.
All that the user interface designer has available in graphical, screen-based interfaces are related to
perceived affordances (Neris & Baranauskas, 2010). Affordance is not the simple presence of an element on
screen, but a suggestion or a clue about how to use it. Figure 1b illustrates a perceived affordance
highlighting a donation button in a web interface. As Norman (2008) explains: [] To Gibson, affordances
did not have to be perceivable or even knowable they simply existed. When I introduced the term into
Table 1. Synthesis of the evolution of HCI field and the concept of affordance

60-70s 80s 90s 2000s 2010s ...
HCI Paradigm

Mainframes Personal computers Network

Ubiquitous computing,
Web 2.0, VR, AR,
Interaction based on
gestures, sensors, 3D
Construction of

Users were
Pragmatic approach.
Ignore it unless it
causes a problem
Meaning interpretation in terms
of information flows
It is the focus,
constructed on the fly,
collaboratively, different
contexts. Machines
replacing humans

requirements (e.g.
Human Factors
Cognitive science
Embodied interaction,
Meaning making
Main question
Computer graphics
How to optimize users
interaction with the

How users might interact
with each other?

How to address human
values into research and

Models and

Systematic methods of testing
Situated action, Distributed Cognition,
Activity Theory, Ethno methodology,
Qualitative approach, Action Theory
Emotion, Aesthetics,
focus on experience
Concept of
Ecological approach,
independent of cognition
(Gibson, 1977)
Applied to
design, part of
Can be associated
with perception
(Gaver, 1991)
Social affordance,
cultural context
(Stamper, 2001)
Social signifier
Motivational affordance
(Zhang, 2008)
(Harrison et al., 2007);
(Harper et al., 2008);
(Bdker, 2006);
(ITEA, 2009);
(Sellen, 2009);
IADIS International Conference WWW/Internet 2010
design in 1988 I was referring to perceivable affordances. [].. Other distinctions between these
approaches are listed in Table 2 (McGrenere & Ho, 2009).
Norman brought the concept to HCI in the
transition of the 1
and 2
paradigms, when the
influence of human factors (and cognitive
psychology) and the goal to fit the relationship
between the human and computer were dominant.
The example of Figure 1c shows that in the 3

paradigm, when technologies such as augmented and
virtual reality are present, the distinction between
physical and perceived affordances is not so clear
anymore. In this example, an interface may be
projected anywhere, but the affordance to touch the
dial with one finger is preserved.
Since Normans view, Hartson (2003), Vyas et al. (2006), Gaver (1991), Stamper (2001), Zhang (2006)
among others considered cognitive aspects in their understanding of affordance and continued to expand the
concept including human aspects such as experience, knowledge, culture, and the social characteristic.
Gaver (1991), for instance, proposes that a combination of affordance with the perceptual information a
person has about it suggests potentials for action, making interaction easy to learn and use. According to him,
affordances are properties of the world that are compatible with and relevant for the actors interaction,
which, when perceptible, offer a link between the actors perception and action.. Figure 1d is an example of
an affordance oriented by a particular situation: a washing machine was used to build a dog house, an
unusual application that would hardly be afforded without an external motivation. Gavers proposal is also in
the transition between the 1
and 2
paradigms, still applied to optimize the humancomputer interaction.
After years of confusion and misuse of the term by designers as Norman himself states (Norman, 1999),
in 2008 he suggested replacing affordance with the term signifier. He argues that the perceivable part of an
affordance is a signifier, and if deliberately placed by a designer, it is a social signifier and asks to forget
the term affordances: what people need, and what design must provide, are signifiers. Because most actions
we do are social, the most important class of these are social signifiers. [] Social signifiers replace
affordances, for they are broader and richer, allowing for accidental signifiers as well as deliberate ones,
and even for items that signify by their absence.. To exemplify, Norman (2008) describes the situation
where the absence of people on a train platform may be a social signifier indicating that the train has already
left. The social signifier includes culture and experiences, similarly to Stampers social affordance idea.

3.2 Stampers Social Affordance
To Stamper (2004), All organisms, including human agents construct their perceptions of the only world
they can know through their actions; they have to discover (or be taught, or inherit by instinct) what
invariant repertoires of behaviour the world affords them (= the affordances); then they populate their
reality with those affordances that help them to survive. Stamper associates the physical affordances with
Gibson's definition linked to properties of the physical environment. They are social in nature, because they
are dependent on the knowledge that has been built up and handed down from generation to generation in a
society. Social affordances are repertories of behavior tuned to the social environment, valid for a certain
community, with a start and finish time, and a starting and finishing authority (Gazendam & Liu, 2005).
Table 2. Comparison of Gibsons x Normans affordances
Gibsons Affordance Normans Affordance
Offerings or action possibilities in
the environment in relation to the
action capabilities of an actor
Perceived properties that may
or may not actually exist
Suggestions or clues as to
how to use the properties
Independent of the actors
experience, knowledge, culture,
or ability to perceive
Can be dependent on the
experience, knowledge, or
culture of the actor
Existence is binary it exists or it
does not exist
Can make an action difficult
or easy
Figure 1a. Physical
affordance of a portable
Figure 1b. Perceived
affordance: highlight of a
button (
Figure 1c. Ex. of a gesture and
augmented reality UI (MIT
Media Lab, 2010)
Figure 1d. Example of
specific perceived
affordance (Street use,
ISBN: 978-972-8939-25-0 2010 IADIS
The photos in Figure 2 illustrate the idea of a social affordance. Once a
person is seen pointing to a map as in Fig 2a, it is clear that this person is
localizing himself/herself or indicating to someone else a specific point in
the map; on the other hand, when a person is pointing to a map with an
electronic device (Fig 2b), it is not clear what the social affordance is
while this behavior is not part of the repertory of the community yet. In
this specific situation, one could guess that he is taking a picture, but he
could be (actually is) using an augmented reality application on the map.
Social affordance is also related to the idea of how an artifact can
stimulate individual or group usage. Figure 3 is a picture of people using an
augmented reality application, which displays on the mobile overlaid information related to the printed map.
Morisson et al. (2009) compared this solution based on the physical map with another application running
strictly on the electronic device and verified that the physical solution had a great potential to group people
and to engage a collaborative work due to the social affordance of the big printed
Stampers Social Affordance and Normans Social signifiers appeared at
different times, but both under the 2
paradigm influence, when culture,
context, experience and life in society started to be taken into consideration by
3.3 Affordance and Motivation
Motivation is an important aspect of human being, when analyzing perception and action; it explains what
causes behavior and why behavior varies in its intensity (Zhang, 2008). Considering that, Zhang (2008)
suggested the term motivational affordance that comprises the properties of an object that determine whether
and how it can support ones motivational needs. She classified some of the most important humans
motivational sources and needs to be considered in design to evoke motivational affordances: a)
Psychological and social needs: autonomy and the self, which refers to the representation of self-identity;
relatedness, leadership and followship, considering social and community life. b) Cognitive motives (beliefs
and expectations): in respect to the users competence and opportunities for achievement. c) Affects and
emotions, which orchestrate human reaction.
The Piano Stair (Volkswagem, 2009) shown in Figure 4 is an example
of exploring humans motivation to stimulate a behavior. By playing
sounds like piano keys, people feel invited to create their own song, which
involves challenge, identity and also social interaction. These features
attract people to choose this stair instead of the escalator most of the
Involving social, cultural and personal aspects together with affect on
interaction, Zhangs definition of affordance goes beyond the 2

paradigm, and may be important to address concepts related to the 3
3.4 Relating Different Approaches
Gibson (1979) first defined affordance as a property of the relationship between the actor and the object.
Gaver (1991) concurs with it, but he disconnects affordances from the perceptual information the person has
about them. Both Norman (1988), with his concept of physical and perceived affordance, and Zhang (2008),
consider affordances as properties of the object. To Stamper (2001) and Norman, with his concept of social
signifier (2008), the social affordances are part of the culture of a society.
Aiming at understanding how these different concepts may be useful in design and taking into account the
current HCI needs and paradigms, an observational study was conducted as described in the next section.
Figure 2a. and 2b. Social affordance
in pointing to a map
Figure 3. Collaborative
use of an AR application
Figure 4. Motivational affordances
in the Piano stair
IADIS International Conference WWW/Internet 2010
As part of an activity in an HCI class, 26 subjects were invited to identify the affordances of two different
artifacts: a Tablet-PC (in tablet mode) and the Apple iPad,
both known but still uncommon artifacts. The study happened
few days after the launch day of the iPad in the United States,
providing a single opportunity to observe the first time
interaction with the device. The students were already familiar
with Normans and Gibsons concept of affordance.
The classroom was split into 2 groups: the first one with 14
students analyzed the iPad first; and the second group with 12
students analyzed first the Tablet-PC.
Both groups first explored individually the artifact, passing
it to each other in the same group; then they were asked to
write down an adjective that expressed their first impression
about the artifact. The first impressions were collected only for
the first analyzed device to avoid a comparative opinion. After exploring the first device they interact with,
they were asked to identify affordances classifying them according to the criteria described in Table 3. In the
second moment, they changed the devices to repeat the analysis and to identify affordances of the second
4.1 Preliminary Results
The identified affordances of both groups were compiled according to its category. A summary of the most
frequent mentions is presented in Table 4 ordered by number of occurrence and illustrated in Figure 5.
Negative aspects are preceded by the symbol (-).
Clearly, the students considered other devices as reference in their first interaction. Both artifacts were
compared with a traditional Laptop ou PC (
) and the iPad with a mobile phone (
) referred to as a big
iPhone by some students on identifyind affordances and also on declaring their first impressions (i.e., big,
light and agile for the iPad and heavy, malleable, and flexible for the Tablet-PC).
The interaction with the iPad created expectations and curiosity. It may explain why the iPad had a
stronger influence in the Tablet PC analysis than the opposite (
). Some affordances of the Tablet-PC were
observed to be more frequent when the iPad was analyzed first. This fact, in addition to the number of
identified affordances of the Tablet-PC that explicitly compared it with the iPad, was not noticed on the iPad
analysis when the Tablet-PC was analyzed first.
Table 3. Criteria to observe the four types
of affordances
Type of
Criteria to observe
Physical Physical properties of the object
Perceived Mental representation, design
Social Related to life in society
Satisfaction of motivational needs:
autonomy, identity, challenge,
feedback, social relations, to influence
and to be influenced by someone,
ISBN: 978-972-8939-25-0 2010 IADIS
The novelty aspect of iPad was predominant in declaring the first impressions of it, and most adjectives
could be classified as motivational or social (strange, stylish, fresh, modern, discovery, motivating, funny). It
is interesting to notice that the novelty sometimes was referred to as strangeness, which can be explained
by the fact that both artifacts were taken as a working tool by some participants(
4.2 Discussion
Although they have some similarities, such as the multi-touch mechanism, these artifacts have completely
different purpose: the tablet-PC is a laptop with some drawing functionality, while the iPad is neither a laptop
nor a mobile phone, it has a new concept originally designed for ebook reading and experiencing the Web.

Figure 5. Examples of identified affordances in the exploratory study
This physical affordance of the iPad clearly influenced the identification of the perceived and
motivational affordances (which includes expectations): because it is bigger than the iPhone, people tend to
expect more features on it considered a motivational affordance in a negative way. On the other hand, the
aesthetic of iPad was the first most present positive motivational affordance and the natural interaction
(feedback of moviment) observed in the perceived affordances also influenced the motivational aspects.
The Tablet PC was more associated to their pen based experience (from drawing and writing on paper).
Seen as a working tool, it was referred to as a laptop with something else. Although considered heavy also
on first impressions declarations, aspects of fragility appeared as physical affordances. Physical aspects
Table 4. Summary of identified affordances
Affordance iPAD Tablet PC

Touch screen (Fig 5a), hold with both hands (Fig 5b),
fragile, weight that suggests strength/value/complexity,
look for buttons
, handwriting
, big
, button sensor or
press? (Fig 5c), different
Pen, on table using
(Fig 5d), heavy
, swivel screen, like a
, excessive number of buttons
, drawing (Fig 5e), less
light and practical than iPad
, save space hiding the keyboard
cannot be used like a book or notebook
Apples mental model, unique button (Fig 5c), inviting
icons (Fig 5f), lack of information
, more options than

because it is bigger

missing references or
interaction standards
, feedback of movements, look for
a keyboard
photos (Fig 5g).
(Fig 5e),

size of icons does not suggest
touch screen
, notes in presentations
, screen swivel only to one
side (Fig 5h), work
, cannot be touched with another objects
(Fig 5i), less strange compared with PCs

Multiple users, symbol of social status (and segregation),
more powerful than iPhone
and more restrictive than a
, inhibitor to be used on public (fear of being stolen),
screen rotation to share content (like a paper) (Fig 5b),
share personal stuff (Fig 5g).
Using in classroom
, its physical aspects do not invite to a
shared interaction - although it is bigger than iPad
, swivel
screen allows to share information (Fig 5h), experience with
pen and touch screen, can be touched by multiple users
(Fig 5j), work/study
, photos/videos/music like iPad

(-)indifference, portable
, toy, novelty, Apples prestige,
(-)missing control of applications and operating

challenging when discovering its
functionalities, motivates touch screen (Fig 5f), showing
personal stuff (Fig 5g), (-)limited to some tasks
, (-)no
software to communicate
Good to draw (Fig 5e), work activities
, (-)artificial writing, (-)
frustration: its not possible to use on the table as a tablet
because the angle of view, a little bit more than a laptop
, touch
screen makes it different
, pen attracts curiosity
, fragile, (-)
no-Linux friendly
, good to avoid repetitive stress injury
IADIS International Conference WWW/Internet 2010
created limitations on social affordances: even being bigger than the iPad, it does not suggest sharing
interaction, mainly because of the angle of screen viewing that is restrictive. The size of icons suggested the
use of the pen in the laptop mode (instead of tablet mode) on perceived affordances. The subjects also
considered it less strange than the iPad taking into account its similarity with a tradicional laptop.
The social affordance of iPad was mostly associated with the brand and social status. Some physical
aspects reinforce the idea of a valuable device. Recent phenomena like hyper-connectivity and techno-
dependency, which can be still more present into the group of participants Computer Science students ,
were taken into account on social affordance and, once again, reflected on motivational affordances.
Besides drawing, the tablets social and motivational affordances are repeatedly related to its formal and
work-related uses in classrooms, presentations and in taking notes. Relating it with the declared first
impressions, the device can be seen as a more versatile laptop.
The identified motivational affordances suggested that this concept should not be taken in isolation; it
may have direct consequences of the other three concepts of affordance, as exemplified below:
Physical: related to attractiveness and affection. Aesthetic, flexibility, portability (weight and size);
Perceived: associated to challenging, feedback: Invitation to use touch screen (or the pen), natural
interaction, friendly user interface.
Social: related to identity, autonomy and social relations: prestige associated to the device or the
brand, identification of being a technical or digital arts working tool, sharing or not information.
The current trends like using portable devices and ubiquitous technologies highlights the importance of
physical affordances that the desktops do not have. Aspects like how to handle, carry or manipulate artifacts
are critical to the technology acceptation and appropriation. The iPhone 4

is an example of the importance

to consider the physical affordance on design process. Holding the iPhone at the bottom left-hand corner
causes signal reception problem. This constraint frustrated users and compelled Apple to offer a special case
and to advise users to hold it in a different way.
According to this analysis, the four different views of affordances were seen as important and
complementary to understand and evaluate the first impression a person may have on exploring a new
technology. Physical and social affordances are central to understand the main aspects that were noticed and
perceived by the person, and as Sellen (2009) states, to understand several levels of interactions. The
perceived affordance is a warranty of understandability of the user interface. And the motivational
affordances can help aggregating them, making it possible to understand how these affordances can
contribute to interaction being part of each persons motivations and social interaction in the 3
paradigm .
The 3
HCI paradigm brings a set of new challenges to the field, such as aggregating human values to the
user interface, but some aspects of the 1
and 2
paradigms still need to be considered when designing
interactions. The perceived affordance is strongly related to some aspects of the 1
and 2
HCI paradigms
that must be preserved in designing interaction that makes sense to its target audience, creating technologies
that are easy to use and easy to learn. The concept of motivational affordance seems to be suitable to the 3
HCI paradigm, once it is associated with human values and their desires and concerns, in a social, economic,
and political ecology (Sellen, 2009).
The concept of affordance has been applied to the interaction design in different ways since Normans
first appropriation, and it has been transformed according to the technical and social contexts of using
technology in daily life. The observational study made it clear that different views of affordance may co-
exist, such as physical, perceived, social, and motivational affordances, and their correlation can be
especially useful to understand how the main aspects of a technology can be associated to each humans
motivation, an important mechanisms that drives humans choices.
In continuity to this work, we intend to focus on motivational affordances, the way they correlate to the
others, how they might explain humans affective reactions, impacting on interface design and evaluation.

ISBN: 978-972-8939-25-0 2010 IADIS
Bdker, S. 2006. When second wave HCI meets third wave challenges. Proc. of the 4th Nordic Conference on Human-
Computer interaction: Changing Roles. ACM, New York, NY, 1-8.
Gaver, W. 1991. Technology affordances. Proc. of the SIGCHI Conference on Human Factors in Computing Systems:
Reaching Through Technology. CHI '91. ACM, New York, NY, 79-84.
Gazendam, M., Liu, K. 2005. The evolution of organisational semiotics: A brief review of the contribution of Ronald
Stamper. Studies in organisational semiotics. Dordrecht. Kluwer Academic Publishers.
Gibson, J.J. 1979. The Ecological Approach to Visual Perception. Boston: Houghton Mifflin.
Harper, R. et al. 2008. Being Human: Human-Computer Interaction in the Year 2020. Microsoft Research.
Harrison, S. et al. 2007. The Three Paradigms of HCI. Extended Abstracts CHI 2007.
Hartson, H. 2003. Cognitive, physical, sensory, and functional affordances in interaction design. Behaviour & IT , 22(5).
ITEA. Roadmap for Software-Intensive Systems and Services, 3
edition, 2009.
Sellen, A. et al. 2009. Reflecting human values in the digital age. Communications. ACM 52, 3 (Mar. 2009), 58-66.
McGrenere, J., Ho, W. 2000. Affordances: Clarifying and Evolving a Concept. Proc. of Graphics Interface 2000.
MIT Media Lab. 2010. Sixth Sense.
interface-augment-our-world. Accessed in 31/08/2010.
Morrison, A. et al. 2009. Like bees around the hive: a comparative study of a mobile augmented reality map. Proc. of the
27th Int. Conference on Human Factors in Computing Systems CHI '09. ACM, New York, NY, 1889-1898.
Neris, V., Baranauskas, C. 2010. User Interface Design informed by Affordances and Norms Concepts. Proc. of The 12th
Int.Conference on Informatics and Semiotics in Organisations (ICISO 2010).
Norman, D. 1988. The Design of Everyday Things. New York, Doubleday.
Norman, D. 1999. Affordance, conventions, and design. Interactions 6, 3 (May. 1999), 38-43.
Norman, D. 2004. Emotional design: why we love (or hate) everyday things. New York: Basic Books.
Norman, D. 2008. Simplicity is not the answer. Interactions, 15 (5). 45-46.
O'Neill, S. 2008. Interactive media: the semiotics of embodied interaction. London: Springer.
Stamper, R. K. 2001. Organisational semiotics: Informatics without the computer? Information, organisation and
technology: Studies in organisational semiotics. Boston, MA: Kluwer Academic Publishers. 115-171.
Stamper, R., et al. 2004 Semiotic Methods for Enterprise Design and IT Applications. Proc. of the 7th Int. Workshop on
Organisational Semiotics, 190-213.
Street Use. Accessed in 10/08/2010.
Tekkotsu. 2010. Cognitive Robotics. Accessed in 10/08/2010.
Volkswagen .2009. The fun theory. Accessed in 10/08/2010.
Vyas, D., et al. 2006. Affordance in interaction. Proc. of the 13th European Conference on Cognitive Ergonomics: Trust
and Control in Complex Socio-Technical Systems. 92-99.
Zhang, P. 2007. Toward a Positive Design Theory: Principles for Designing Motivating Information and Communication
Technology. Designing Information and Organizations with a Positive Lens. Elsevier.
IADIS International Conference WWW/Internet 2010
Daniel Kappler* and Paul Darbyshire**
*University of Liverpool, Laureate Online Education
*Laureate Online Education, B.V., De Entree 59-67 1101 BH Amsterdam Z.O, The Netherlands
**School of Management and Information Systems, Victoria University
**PO. Box 14428 Melbourne City MC, Melbourne, Victoria, Australia, 8001
Search Engine optimization attempts to optimize website content in order to get to the top position of the Search Engine
Results Page (SERP). Over the years search engine algorithms have had to be changed as spamming attempts such as
keyword stuffing, or the use of link farms eroded the importance of relevant content to favor the content offered by
individuals using these procedures. For enterprises which are operating websites, the promotion of their presence is a
vital part of successful marketing. For the average website owner or manager it seems very difficult to decide what
procedures have to be applied to make a websites presence more successful. Out-dated procedures are still widely used,
so enterprises have a very low chance to reach the Internet users through standard searches who are looking for the
products or services they offer. This paper discusses the development of a model to facilitate non-IT professionals (in
particular SME's), in gaining an understanding of the details of Search Engine Optimization and equip them with the
skills for performing these tasks. The model is constructed with feedback loops to allow for the ongoing optimization
required in a continual changing environment. We evaluate the model with three company websites which will be
analyzed and recommendations will be given to improve the content and source code of selected pages. The website
owners or managers will be given instructions to implement the recommended optimization over a specific period. The
feedback loops will help show if the suggested procedures have been implemented, and how these were handled.
Search Engine Optimization, Model, SME's. Keyword Search
Within the last decade the way that consumers have been searching for products and services has changed
profoundly. For a typical online shop, operating a website has been always essential, but not for enterprises
which have been used to acquiring new customers in the traditional way (Georgiou.and Stefaneas 2002). In
highly developed technology areas of the world the PC is the main instrument for finding information. The
big players have learned to cope with the changing search engine algorithms and made their websites visible,
but smaller enterprises are often left behind. Yet, he average website owner believe that having an online
presence is enough to attract new customers. There is still a common misconception that it is sufficient to
have a few Meta tags in the source code which describe the companys products and services., however, that
there are still an alarming number of web pages which do not even have a title.
For companies who want to improve their websites it is often difficult to know where to start. Often
dubious Search Engine Optimization(SEO) providers take advantage of the website owners lack of basic
knowledge. There have been cases where enterprises bought an SEO package, hoping to get a top ranking on
the Search Engine Results Page (SERP) of the major search engines. This might only work for a short period,
until search engine algorithms recognize the spamming attempts, and then punish the site by downgrading its
rank, or in worst case banning the site from the indexes at all (Ross 2008). Generally, the website owners or
managers who are non-IT professional need to be provided with clear step by step instructions how to
optimize their web presence. An enterprise which has neglected the optimization of its website first needs to
ISBN: 978-972-8939-25-0 2010 IADIS
recognize the necessity of improvement. Once the enterprise is aware of the potentials it needs to be
instructed how SEO best practices have to be implemented.
In the following sections we discuss the development of a model to aid in the facilitation of non-IT
professionals (in particular SME's), in gaining an understanding of the details of Search Engine Optimization
and equip them with the skills for performing these tasks. This will help improve the overall penetration of
their Web sites. The model is constructed with feedback loops to allow for the ongoing optimization required
in a continual changing environment. We evaluate the model by applying its techniques to three selected
cases where an optimization promises to improve a website's ranking, and therefore increase the website's
traffic. Evaluating the results will show if the implemented improvements will lead to a more successful web
presence of these selected enterprises. It is equally important to find out if procedures for optimizing websites
can be reproduced, and if they can be applied the same way in the near future. These evolving technologies
will be also critically evaluated.
Webmasters, developers, programmers and SEO experts have been learning to live with the way search
engines absorb and present the available information on Internet, but the website owner often remains
disoriented. Usually website owners and webmasters have to work together to make an enterprise's website
visible. Scanlan (2008) clearly demonstrates that when someone searches for products or services and your
site does not appear on the first page of search results, than your site is considered to be invisible. This is a
harsh reality as usually there are only 10 places available per page. Ghose and Yang (2008) declare that
SEO refers to the process of tailoring a website to optimize its unpaid (or organic) ranking for a given set
of keywords or phrases. Organic optimization means that no advertisement campaigns have been used. The
success depends fully on the optimization of the content, site structure, and external and inbound links.
Generally, non-organic optimization falls under the definition of Search Engine Marketing (SEM). Search
Engine Optimization is all about making the pages of a website visible for search engine spiders. To start
with the optimization the actual performance of the website has to be analyzed. Important factors are the
number of indexed pages, inbound links and PageRank. The latter factor depends on the popularity and age
of the pages. Scanlan (2008) recommended that the goal of the optimization is to be at least two units ahead
of the PageRank of your closest competitor. According to Brin and Page (1998) PageRank was a new
approach how the web pages were ranked.
Much of Search Engine Optimization is centered on the selection of popular keywords and phrases.
Usually the web page content already provides keywords which can be used for optimization. Once a list of
keyword has been created it can be expanded by brainstorming. King (2008, p.13) recommended
benchmarking competitors websites to see which keywords they use, and also to include plurals, splits,
stems, synonyms and common misspellings. For the website owner or manager there are various free online
tools available for the keyword research. King (2008) mentions that these free tools are limited compared to
paid services like Wordtracker. For ambitious website operators and SEO professionals he recommended
using Wordtracker for streamlining the process of key phrase discovery. The ultimate goal in SEO is to
discover the primary key phrase that accurately describes the overall business, product, or service but which
still has adequate search demand (King 2008, p.15).
Keywords which are placed in the URL and title of a web page have the highest impact on search engine
spiders. Also keywords should be also included in the Meta section of the source code. Experience has shown
that keywords placed in Meta description have more value then those in the Meta keywords. The reason for
this is that in recent years many webmasters were abusing this section of the source code by keyword
stuffing. Therefore search engine algorithms had to be changed to disregard such spamming attempts. Search
engines today don't put too much emphasis on the keyword Meta tag, as it is largely ignored. Therefore
critical keywords and phrases should embedded within bold tags and heading tags. Also the use of keywords
is recommended in anchor text and hyperlinks. Using these tags puts more emphasizes on the selected
keywords and phrases, and improves the chance to be indexed by search engines.(Rognerud 2008)
While, SEO is largely focused on the selection of keywords, and link building, any gains in these areas
can be degraded by an unfavorable site design, and overloaded source code. A site design which is user
friendly is not always search engine robot friendly. Graphical enhanced websites using JavaScript or Flash
IADIS International Conference WWW/Internet 2010
technology can be a barrier for search engine spiders. In such cases challenge is then to change the source
code to make the site more discoverable (Walter 2008).. The optimization of the source code is usually
difficult to understand by a website owner or manager who is not an IT professional. One way to check if a
site structure does not oppose any barriers to search engines is by disabling the JavaScript and CSS functions
to see if all content is still available. When designing or optimizing a website a progressive enhancement is
recommended. By keeping the structure (HTML), presentation (CSS), and behavior (JavaScript) separate the
content remains accessible to search engine spiders (Walter 2008).
If an enterprise hires a SEO specialist for optimizing its Web presences, or tries to get the job done by
itself, then it is vital to specify Web metrics. To assist in analyzing online traffic, indexed pages, link
popularity and PageRank, various software packages and online tools are available. With their standardized
reports typically user sessions and details, and the quality of the websites source code can be determined.
For marketing strategies it is important to know how users access the pages, and from where they come. But
some limitations have to be taken into account when interpreting data from server log files. Among these
limitations are cached pages which are not logged by the server, search engine robots requests or the true
location of the visitors. (Weischedel and Huizingh 2006). Even though the website owner has to anticipate
imprecise results, the general trend of the performance should be observed. Some pages might perform better
than others. By monitoring the entering and the exit pages, and keeping an eye how long visitors spend on
each page will give an idea what keeps the visitors interested, and what might cause them to leave your site
(Rognerud 2008).
The model for Search Engine Optimization was developed from the literature and involves the stepwise
process of search engine optimization, to be undertaken by following a number of stages. The actual model
itself consists of five stages which have been adapted from the classic 'Waterfall Model' of software
development (Royce 1970). The classic stages have been renamed somewhat to reflect their roles in the
context of the model, see Figure 1, with descriptions below. The selection of the modified waterfall model
seemed natural as a framework for an SEO model due to the iterative, sequential and compartmentalized
nature of the task. The iterative nature of Search Engine Optimization is very well represented by the
feedback loop contained in the modified waterfall model to that originally proposed by Royce.

Requirements data needs to be collected to find out how SEO has been handled by SMEs, and how
search engines in general weight online content. Here it will be also described what tools and
software will be required to achieve the objectives.
Specification the competition and the actual current state of the website performance will be
analyzed before the objectives can be specified. Here the keywords matching the content will be
specified which will be the basis for a successful implementation.

Figure 1. Adapted Waterfall Model for SEO
ISBN: 978-972-8939-25-0 2010 IADIS
Source Code Optimization This stage incorporates the implementation of the defined keywords
and phrases from the previous stage, and the widely neglected optimization of the source code
effectiveness. The same program which was used for defining the enterprises market position will
be also useful for analyzing the website performance with regard to source code effectiveness and
eliminating errors
Offsite Optimization At this stage of the optimization the persons involved in the process will be
asked to actively popularize the presence oft their website. This will be achieved by bookmarking
the web pages in various social bookmark services, registering the site to online indexes, and
requests to webmasters of partner-sites to place back links.
Testing To measure the effectiveness of the suggested implementation website operators need to
be provided with tools. To facilitate the introduction of testing the performance, these tools need to
be readily obtainable. In this case it is recommend using the Web CEO program, and selected online
services. Before and after the optimization the same tools will be used to receive results which will
be based on the same measurement.
The data required to start the optimization process will be provided by a preliminary questionnaire which
is supplied to company owners and managers who are interested in optimizing their websites. This data will
be necessary to analyze how the Search Engine Optimization has been handled in the past, if any, and if the
website operators are aware of the importance of their online presence and competition. To measure the
effectiveness of the suggested implementation website operators need to be provided with tools. To facilitate
the introduction of testing the performance, these tools need to be readily obtainable. In this case it is
recommend using the Web CEO program, and selected online services. Before and after the optimization the
same tools will be used to receive results which will be based on the same measurement.

Figure 2 Shows the high-level flow of the steps in application of the the model, indicating that the model
is an iterative one which reflects the nature of the SEO process. Thus optimization may need to be performed
again over time. The feedback loop in Figure 2 from the final stages of acceptance testing and evaluation of
the recent optimizations, feed back into the initial performance testing before optimization.
The model was tested against three SME websites which showed a possibility for improvement during the
optimization phase. The SME's chosen were similar companies in that they all dealt in language training.
Keyword research
and selection
SEO Inventory
Performance test
Optimizing source
Acceptance test
and evaluation


Figure 2. Iterative nature of the Model and Optimization Process
IADIS International Conference WWW/Internet 2010
FSTS Sprachreisen (Language Trip Courses). The website, had been online since May
2000. The website already had a Google PageRank of 5.
Deutschinstitut (German Language Institute). The company website is which
has been online since May January 2004. The website had a Google PageRank of 4.
SALT Pro Skills and Language Training. The companys website is at website
and has been online for 2 years. The website had a Google PageRank of 3
All three companies had a potential for improvement by applying optimization techniques. The detailed
steps taken during the optimization process, by applying the Model discussed in the previous section were:
Installation of the Tools: The website owners were instructed how to install Google Analytics and Web
CEO. At the end of the SEO Inventory stage, the website operators were required to answer questions
regarding the installation process. These questions were be included in the SEO Inventory feedback.
SEO Inventory: For monitoring the progress of the optimization and noting changes in the website
performance, basic data was be required: number of visitors per day; average time visitors spend to look at
pages; percentage of new accesses; percentage of visits through search engines; number of indexed pages;
number of inbound links; PageRank.
SEO Inventory Feedback: It was important to learn if website owners or managers were able to install the
required programs and tools to benchmark their website, and compare the performance to their competitors.
Additionally, it was a learning process in how to deal with the SEO process.
Keyword Research: The website operators were instructed how to research for suitable keywords. The
first step was to select the appropriate tools which gave valuable information about a website's current state
regarding the keyword selection. For the 3 case studies different online services and the program Web CEO
were used. One important factor of a successful optimization is the density of keywords used on a web page.
Among web designers and SEO experts a value of 2 to 5 percent is accepted. If the density of a keyword is
below 2 percent of the content, then web page does not seem relevant enough to be listed on a Search Engine
Results Page (SERP). A value over 5 percent can be already classified as spam attempt, and be downgraded
by search engines.
Analyzing Competitors: After popular keywords and phrases have been defined, the website operators
have to define their closest competitors. Here the enterprises will be confronted with the reality. It is
important that popular keyword combinations will be tested for determining the web pages ranking on the
Search Engine Results Page (SERP).
Selecting and Placing the Keywords: Website owners or managers learned the importance of selecting
and placing the appropriate keyword in the pages content. In some cases the website operator was not be able
to edit the page source code which was necessary to optimize the title and Meta tags. In such cases the task
was given to an IT professional who was instructed to edit the section of the page(s).
Finalizing the Keyword Research and Implementation Stage : The process will be evaluated after
recommendations were given to the website owners or managers how to use appropriate keywords to
improve the SERP. The participants will have to answer questions like:
what were the prominent keywords on the main page
what online tools were used to find popular keywords and phrases
what programs and tools were used to benchmark their competitors websites
what pages of their website were optimized
what keywords and phrase were used for optimization
how keywords were emphasized in the pages source code
how many pages were optimized
who participated in the keyword research
who implemented the keyword optimization
Optimizing Source Code: An optimized source code is one of the major preconditions for being indexed
by search engine spiders, and to get a good ranking on a Search Engine Results Page (SERP). For the non-IT
professional not familiar with (X)HTML it can be difficult to identify source code which could be optimized.
The selection and placing of the keywords might be already challenging for the average website owner or
manager. For that reason the cooperation with IT professionals is needed. The person who designed and
programmed the website has to analyse, and if necessary optimize the source code.
ISBN: 978-972-8939-25-0 2010 IADIS
The website owner can use a check-list to verify everything possible has been done to make sure
that all the pages have been made visible for search engine spiders, and load faster. [20] This
will include that:
everything unnecessary has been removed like gif, java or flash animations
that there are no nested tables (browsers need more time to parse them and search engine robots
are limited to find relevant information)
CSS (Cascading Style Sheets) are used to separate the content from the presentation
images are optimized by reducing their file size, and inform the browser about their dimensions
the code has be cleaned up by removing unnecessary spaces between tags, empty tags and
comments in the source code
the cleaned-up code has been validated to meet W3C recommendations
There are programs and online tools available, so the website owner can compare his or her website to
competitor websites. The following programs and online tools will be incorporated in the optimization
process: Web CEO,, and
Locating further Sources of Errors: It is important that further errors can be located. To achieve this it is
recommended to use a computer program like Web CEO which shows more details rather than an online
service. According to Web CEO Auditor function the code optimization will be focused on:
errors: problem pages; broken internal links; broken external links; broken anchors
usability: page size; slow pages; deep pages
maintenance: redirects; missing / invalid titles and meta data; new pages ; old pages; small pages;
missing ALT attributes or images; missing WIDTH / HEIGHT attributes of images; orphaned files
The website owner and his or her webmaster should be able to get a report about the site quality. The
objective of this procedure will be that the criticized points need to be fixed by the person who has the
appropriate knowledge and access to the source code, either the website owner or webmaster. For
benchmarking the website and those of the closest competitors, o