improving learning management through semantic web and social networks in e-learning environments

9
Improving learning management through semantic web and social networks in e-learning environments M.P. Cuéllar , M. Delgado, M.C. Pegalajar Department of Computer Science and Artificial Intelligence, Escuela Técnica Superior de Ingeniería Informática y Telecomunicaciones, C/. Pdta Daniel Saucedo Aranda s.n., University of Granada, Granada, Spain article info Keywords: Social networks Learning Management Systems e-Learning Ontologies abstract Internet social networks have arisen in the last years as powerful tools where people exchange knowledge and multimedia content. They help to share interests between groups of people with common features. Undoubtedly, there is an inherent social network in any e-learning system, where the main actors are teachers, learners and learning resources. Most e-learning software are mainly focused in content dissemination and group work, but the possibilities that Internet LMSs could offer go further. Recently, there has been research work focused on Web Communities for learning and their formulation as Social Networks. Thus, social network analysis may be applied to infer group structures and to make intelligent recommendation systems and data mining. This paper proposes a method for the formulation and interpretation of learning management plat- forms as social networks. In order to achieve a major generalization, we develop an ontology to integrate the information from different Learning Management Systems. After that, a personalized social network is extracted from the ontology. This change in the point of view of a LMS could be a challenge to make further studies about learners, teachers and learning resources to obtain a better understanding of their social structure, and therefore to make or improve decisions about the learning process. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction 1.1. Aims and scope The advances in Web 2.0 and XML-based technologies are changing our concept about WWW by mean of the inclusion of semantics in the web. Accordingly, e-learning systems are also benefit of the semantic web (Nilsson, Palmr, & Naeve, 2002). Most of efforts in this field share the use of ontologies to provide learn- ing material with semantics (Dunkel, Bruns, & Ossowski, 2006; Dzbor, Stutt, Motta, & Collins, 2007; Yli-Luoma et al., 2006), often included within a service-oriented or multi-agent architecture (Dietze, Gugliotta, & Domingue, 2007; Dunkel et al., 2006; Henze, 2005b). The goals pursued by these proposals are wide: For exam- ple, the work Huang, Webster, Wood, and Ishaya (2006) proposes a four-stage method to improve self-learning, taking account an analysis of learner personality. In Henze (2005a, 2005b), the aim is to improve the workspace personalization using a service-ori- ented architecture and RDF/S. The proposal in Jovanovic ´ et al. (2007) offers a model to provide teachers with feedback informa- tion about students and learning resources interaction. In Dunkel et al. (2006), a multi-agent system is developed to provide students with intelligent recommendation for their tasks. Other proposals go further and suggest the study of semantic social interactions in e-learning (Torniai, Jovanovic, Gasevic, Bateman, & Hatala, 2008). In this work, our approach is closed to this idea. Our aim is to provide a formulation of any learning management platform as a social network, in order to be able to do social network analysis (SNA) over teachers, learners, learning resources and their interac- tions. The information within these systems is usually stored in a relational database. Then, for a particular case, the problem would be to identify actors and relations in the database tables and to transform these tables into a social network. However, the main gap found in this idea is the loss of generality: database designs may differ deeply depending on the purpose and requisites of the system. Thus, in order to fulfill SNA over the information from mul- tiple LMSs, it would be hard to match all databases with their cor- responding social networks and, moreover, to integrate these networks for a general analysis. As a previous step, it should be provided a common framework where the storage design of actors and relations, and the concepts within an e-learning system, are independent. To solve this, we propose to match the relational database structure of the LMSs with an ontology. The ontology proposed has an initial structure with classes and relations, but it could be extended by an expert to fit the database 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2010.09.080 Corresponding author. Tel.: +34 956 52 61 59. E-mail addresses: [email protected] (M.P. Cuéllar), [email protected] (M. Delgado).. Expert Systems with Applications 38 (2011) 4181–4189 Contents lists available at ScienceDirect Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa

Upload: mp-cuellar

Post on 26-Jun-2016

214 views

Category:

Documents


2 download

TRANSCRIPT

Expert Systems with Applications 38 (2011) 4181–4189

Contents lists available at ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier .com/locate /eswa

Improving learning management through semantic web and social networksin e-learning environments

M.P. Cuéllar ⇑, M. Delgado, M.C. PegalajarDepartment of Computer Science and Artificial Intelligence, Escuela Técnica Superior de Ingeniería Informática y Telecomunicaciones, C/. Pdta Daniel Saucedo Aranda s.n.,University of Granada, Granada, Spain

a r t i c l e i n f o

Keywords:Social networksLearning Management Systemse-LearningOntologies

0957-4174/$ - see front matter � 2010 Elsevier Ltd. Adoi:10.1016/j.eswa.2010.09.080

⇑ Corresponding author. Tel.: +34 956 52 61 59.E-mail addresses: [email protected] (M.P.

(M. Delgado)..

a b s t r a c t

Internet social networks have arisen in the last years as powerful tools where people exchangeknowledge and multimedia content. They help to share interests between groups of people with commonfeatures. Undoubtedly, there is an inherent social network in any e-learning system, where the mainactors are teachers, learners and learning resources. Most e-learning software are mainly focused incontent dissemination and group work, but the possibilities that Internet LMSs could offer go further.Recently, there has been research work focused on Web Communities for learning and their formulationas Social Networks. Thus, social network analysis may be applied to infer group structures and to makeintelligent recommendation systems and data mining.

This paper proposes a method for the formulation and interpretation of learning management plat-forms as social networks. In order to achieve a major generalization, we develop an ontology to integratethe information from different Learning Management Systems. After that, a personalized social networkis extracted from the ontology. This change in the point of view of a LMS could be a challenge to makefurther studies about learners, teachers and learning resources to obtain a better understanding of theirsocial structure, and therefore to make or improve decisions about the learning process.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

1.1. Aims and scope

The advances in Web 2.0 and XML-based technologies arechanging our concept about WWW by mean of the inclusion ofsemantics in the web. Accordingly, e-learning systems are alsobenefit of the semantic web (Nilsson, Palmr, & Naeve, 2002). Mostof efforts in this field share the use of ontologies to provide learn-ing material with semantics (Dunkel, Bruns, & Ossowski, 2006;Dzbor, Stutt, Motta, & Collins, 2007; Yli-Luoma et al., 2006), oftenincluded within a service-oriented or multi-agent architecture(Dietze, Gugliotta, & Domingue, 2007; Dunkel et al., 2006; Henze,2005b). The goals pursued by these proposals are wide: For exam-ple, the work Huang, Webster, Wood, and Ishaya (2006) proposes afour-stage method to improve self-learning, taking account ananalysis of learner personality. In Henze (2005a, 2005b), the aimis to improve the workspace personalization using a service-ori-ented architecture and RDF/S. The proposal in Jovanovic et al.(2007) offers a model to provide teachers with feedback informa-tion about students and learning resources interaction. In Dunkel

ll rights reserved.

Cuéllar), [email protected]

et al. (2006), a multi-agent system is developed to provide studentswith intelligent recommendation for their tasks. Other proposalsgo further and suggest the study of semantic social interactionsin e-learning (Torniai, Jovanovic, Gasevic, Bateman, & Hatala,2008).

In this work, our approach is closed to this idea. Our aim is toprovide a formulation of any learning management platform as asocial network, in order to be able to do social network analysis(SNA) over teachers, learners, learning resources and their interac-tions. The information within these systems is usually stored in arelational database. Then, for a particular case, the problem wouldbe to identify actors and relations in the database tables and totransform these tables into a social network. However, the maingap found in this idea is the loss of generality: database designsmay differ deeply depending on the purpose and requisites of thesystem. Thus, in order to fulfill SNA over the information from mul-tiple LMSs, it would be hard to match all databases with their cor-responding social networks and, moreover, to integrate thesenetworks for a general analysis. As a previous step, it should beprovided a common framework where the storage design of actorsand relations, and the concepts within an e-learning system, areindependent. To solve this, we propose to match the relationaldatabase structure of the LMSs with an ontology.

The ontology proposed has an initial structure with classes andrelations, but it could be extended by an expert to fit the database

4182 M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189

structure in order to enrich the knowledge extracted. Thus, the in-stances in the ontology could be automatically extracted from therelational model through the semantic matching. Our main inter-est is to give a common abstract framework to organize usersand learning resources information, but the resulting ontologycould provide many advantages: semantic would be added to enti-ties and relations and therefore LMSs could share semantic infor-mation between them, therefore making it possible to extendlearning capabilities between users of different LMSs. Furthermore,semantic information retrieval techniques could be used to im-prove the system recommendation capabilities and the learningcontent dissemination. But more importantly for our approach isthat the common framework generated would allow us to developa procedure to transform the information in the ontology to a so-cial network structure. Thus, we are able to make social networkanalysis over teachers, learners and learning resources to obtainrelevant information about the social structure inside the LMS,which teachers could use to improve their teaching methods, stu-dents groups or to provide a better personalized teaching. Thischange in the point of view of a LMS could be a challenge to makefurther studies about learners, teachers and learning resources toobtain a better understanding of their social structure, and there-fore to make or improve decisions about the learning process.

1.2. Social networks

The term Network is different depending on the field to study.In social sciences, a social network comprises a set of people orgroups of people (actors) and their interactions (ties) (Newman,2003). The representation of a social network is usually given inthe mathematical form of a graph G = (V,E), where the set of nodesV means the set of actors and the set of edges E # V � V containsthe relations between them (Hanneman, 2005). If an interactione 2 E is labelled with a single value (as for example ‘‘John knowsPeter”) it is said that the social network is simplex. On the otherhand, if e has more than one value (for example, ‘‘John knowsand has the same interests than Peter”) then G is a multigraph andthe social network is named multiplex (Izquierdo & Hanneman,2006).

Social networks have been widely studied since the 20’s in dis-ciplines like sociology (Hanneman, 1988) or economy (Mayer,2009). In the last years, the increase in the use of Internet and userinteractions in the WWW allows computer scientists to use socialnetwork analysis techniques (Ehrlich and Carboni, 2005) for datamining and knowledge discovery in large Internet social networks(Weaver & Morrison, 2008). Examples of popular social networkssites in the Internet are Facebook, MySpace, Tuenti, YouTube andOrkut but, in general terms, any web community and also the blog-sphere could be considered social networks. Usually, these sites of-fer services like list of friends, people surfing, messages, eventsmanagement and media uploads. The future of social networks inthe Internet is promising, and it has been discussed in Breslinand Decker (2007).

Besides the current impact of social networking in user experi-ences, these models are also used to study people relations andprofiles. Enterprises and researchers take advantage of these anal-yses to find new business models and to validate their assumptionsover a case study population. An example is the work Ross et al.(2009) which makes a study of people’s personality through Face-book use. Traditionally, SNA has been focused to obtain informa-tion from properties of the graph that represents the socialnetwork, such as centrality of nodes, walks and distance betweenactors, clustering detection, eccentricity, connectivity, etc. Theseproperties have been mainly studied from the point of view ofgraphs theory and statistics (de Nooy, Mrvar, & Batagelj, 2004;Hanneman, 2005). The analysis may be done using specific soft-

ware for SNA as described in de Nooy et al. (2004), Huisman andvan Duijn (2005), although other usual mathematical environ-ments could be used (Izquierdo & Hanneman, 2006).

Recently there have been research and projects to study socialnetworks across the Web. A general description of some applica-tions may be found in Staab et al. (2005). Most of these effortscombine Semantic Web and social networks in the same way:the use of ontologies to create a common framework for knowl-edge organization and semantic information sharing. For example,reference Zhou et al. (2008) use an ontology to integrate the infor-mation from DBLP and LinkedIn to find relations of collaborationsbetween authors in research works. The article Jung and Euzenat(2007) describes a model to integrate three networks in a virtualcommunity: social network, ontology network and concept net-work. The authors call the resulting net a Semantic Social Network,and show how to infer information in a network from data existingin the other two ones. A system for the integration of social net-works using ontologies, OpenSocial, is described in Mitchell-Wonget al. (2007). The work in Cantador and Castells (2006) describes amethod to identify semantic social networks from an ontology ofuser profiles. The user interests are then studied to make collabo-rative filtering and recommendation models. The proposal in Gan-ley and Lampe (2009) applies SNA over Slashdot, an online socialnetwork software. In this system the users are tagged with a rep-utation value called Karma. The users Karma is included in the net-work analysis to find brokerage, closure and other properties insidethe net. Other example is the article Thushar and Thilagam (2008)which uses RDF to encode a social network and semantics betweenits components. This RDF model is then used to identify semanticassociations inside the network. In Hamasaki et al. (2007), a tripar-tite model to represent ontologies is used to make an integration ofsocial networks about authors and conferences. Recently, a modelto give a representation and a query language for social networkshas been presented in Martn and Gutierrez (2009) using RDF andSPARQ as the main stack for the language.

Other techniques regarding Artificial Intelligence and statisticshave been also proposed to study social networks, besides Seman-tic Web and ontologies. In Zhdanova, Predoiu, Pellegrini, and Fen-sel (2007) is proposed a social network representation withhypergraphs to model a web community. It also provides mathe-matical tools for SNA over the hypergraph, to achieve the goal offinding closeness of community users according to the contentsthey upload. The social structure in the Web has also been studiedby means of Internet hyperlinks exploration in Park (2003), toachieve the goal of effective link data retrieving. In Bhatia and Gaur(2008), is proposed a statistical method for community mining in-side social networks. Makrehchi and Kamel (2006) describes amethod to predict the topology of a large social network when onlya few relations between actors are known. The problem is formu-lated as a text mining problem and it is solved using Support Vec-tor Machines for classification. The same authors applies a multipleresampling method to solve this problem in Makrehchi and Kamel(2007). In Jung, Koo, and Jo (2007) the authors point out that usu-ally there are heterogeneous relations inside social networks, andprovide a Divide-and-Conquer approach to find these relationsover semantic social networks. Neural networks and case-basedreasoning have been applied to build resources recommendationsystems for community users in Kanawati and Malek (2007), andalso data mining techniques have been used to identify social net-works in relational databases (Hensen & Neville, 2003).

Undoubtedly, there is an inherent social network in any collab-orative e-learning Internet software. This network ties betweenteachers, learners and learning material are encoded inside entitiesand relations in the database of the system. It could be useful if ateacher could make SNA over this learning network, to find proper-ties such as closeness between different learners according the

M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189 4183

material they upload/download, abstract student profiles to makea better personalized training, recommendations of learning re-sources, etc. However, database designs may differ deeply fromone database to another. It would be hard to develop a generalmethod to match any database with the social network it encodesautomatically and more importantly, to understand the resultingnetwork, in a semantic sense, to do a proper analysis. Our researchsolves this problem, and it may be described in two stages:

� Firstly, an ontology for e-learning environments is developed.After that, we make a matching between the ontology classesand properties, and relational databases of e-learning Internetsystems. The data from the databases could be imported andsaved as ontology class/property instances. In this step, weobtain a common framework for data sharing between differente-learning systems. Moreover, the database is given withsemantics, which provides the advantages of Semantic Webwe introduced in Section 1.1.� Secondly, the data stored in the ontology is transformed and

exported as a social network. The hierarchical structure of clas-ses and properties in the ontology is here a powerful tool, whichallows us to obtain different social networks for either a finegrained or a high level analysis, depending on the classes andproperties we select for the study.

Section 2 shows the main design of the ontology and frameworkfor social network analysis in e-learning. After that, Section 3 pro-vides a procedure to build the social network from the ontology.Section 4 discusses some classic properties of SNA that could beinteresting to study over a LMS. Section 5 shows an example ofour approach over a LMS. Finally, Section 6 concludes and de-scribes further work.

2. A general framework for social network analysis ine-learning

Social relations within an e-learning system are encoded in-side entities and relations in the relational database of the learn-ing platform. In order to make a proper social analysis overlearners, teachers and learning material for different LMSs, it isrequired a common framework to represent actors and socialrelations. In this section, we propose a domain ontology basedin FOAF (Breslin & Decker, 2007) to achieve this goal. Once thematching is done, a transformation could be applied over theontology structure and the class instances and slots to obtain apersonalized social network containing social relations between

Fig. 1. Ontologies as the abstract common framework for information sharing.

teachers, learners and learning resources. Fig. 1 illustrates thisidea.

The hierarchy of the ontology plays an important role in or-der to provide a personalized social network. We have built asoftware named OntoLMS in order to allow the users to selectproperties and actors for social network analysis, and to providethe social network that best represents the user interests. Ont-oLMS works as a wizard to ease the user interface: First of all,the ontology files are loaded in memory. We have used theOWL API for Java as the parser with the reasoner Pellet. Secondly,the user may specify one or more LMS databases in Internetfrom which data are obtained, and matches the ontology withthe data in all the databases. The matching is carried outsemi-automatically. An expert matches manually the ontologyclasses and properties, since it depends highly on the databasedesign and it would be hard to obtain a suitable automated100% reliable procedure for ontology and database matching.After that, the data from the databases matched are extractedand mapped into ontology class instances and slots. Once thedata is stored in the ontology, the user is allowed to identifythe properties and/or classes of interest for the social networkanalysis. Finally, the social network that best suits the userrequirements is generated and provided to the user. These stepare depicted in Fig. 2. Next Section 2.1 overviews the ontologydesign and shows the hierarchy of classes and relations in depth.Steps 1–4 for the ontology and database matching are reviewedin Section 2.2. In the next sections, we assume that the datafrom the databases are stored in the ontology, and the procedureto obtain the social network (steps 5–6) is described inSection 3.

2.1. Ontology design

The social relations between actors in a social network aretypically represented about people knowing other people. TheFriend of a Friend project (FOAF) is one of the most popularontologies in the web for this purpose (Breslin & Decker, 2007;Staab et al., 2005; Hamasaki et al., 2007). FOAF is a vocabularyfor describing people and relations such as who knows who,but it is not a complete data sharing solution since its vocabularyand properties are general and limited. We have used FOAF asthe starting point to build the ontology, in order to achieve thegoal of standardization, but further classes and properties are re-quired to consider fine grained relations between entities inLMSs.

The extension we propose is foafLMS, a hierarchical organiza-tion of actors and relations in the system, where the concepts Agentand Document are the main abstract representations of entities,and Knows and theme are the main abstract relations. We haveused the software Proté.g.é from the Stanford University to buildthe ontology design. Fig. 3 shows an example of the hierarchicalorganization of the ontology.

The hierarchical organization we propose allows us to study theelements of e-learning environments at different levels of granu-larity. Using this structure, our goal is to bring the possibility to ex-tract personalized social networks depending of the actors andrelations of interest for the analysis. Thus, the abstraction Agentcould be either a Person (a Teacher or a Student) or a Group. Somerelevant groups are a Department (which is a set of teachers), a Sub-ject (a set of teachers and students working in specific issues) or aWorkGroup (a set of people working together). The abstraction Doc-ument encompasses a ResourceForStudy (learning material usuallyprovided by teachers or a work group) and a Work (documentsgenerated by a student or a work group). For further granulariza-tion, there are different specialized learning resources such as

Fig. 2. Procedure for ontology data extraction.

Fig. 3. Snapshot of the extension of the ontology classes (left) and properties (right) structure.

4184 M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189

Notes, Exercises, Media, etc. Additionally, classes Evaluation and Is-sue give support for agent evaluations and association of documenttopics and agent interests, respectively.

Considering the ontology object properties, they have beenalso organized in a hierarchy it in order to provide a personalizedsocial network with our approach. The most abstract relation be-tween agents is knows, and it represents any way in which anagent could be related with another agent. To achieve fine-grained analysis, we may find sub-relations to represent that ateacher teaches a student and its inverse (isStudentOfTeacher), adepartment offersSubject a subject, a person isFriendOf anotherperson, etc. On the other hand, the abstract property theme mod-els relations between Agents and Documents. For specialized rela-tions we find sub-properties like isInterestedIn, studiesIssues, etc.Specific relations between Documents are also modelled as sub-properties of theme, as for example isTheOppositeOf, isTheSameAs,isAPartOf, etc. Finally, other relevant slots make associations be-tween agents and their evaluations (hasEvaluation), and the issueswhich an evaluation is about (itemsEvaluated). Other properties inFOAF could also be used to make associations between classes inthe ontology, such as maker, made, interests, etc. (Brickley & Mill-er, 2007).

2.2. Ontology and database matching

This section reviews the procedure we have used to map thedatabase information into ontology classes and properties (Cuéllaret al., accepted for publication). We use the notation of relationalalgebra since it is a classic language to describe relations in dat-abases and has enough expressivity for our purpose:

� The URI for an individual ci of a class C in the ontology ismatched with a row contained in the natural joinffln

x Tx ¼ T1 ffl T2 ffl � � � ffl Tn of a set of tables {T1,T2, . . . ,Tn} anda set of logic constraints HC over tables attributes. Both HC

and tables {Tx} are provided by an expert. A database identifica-tion D is also necessary in order to ensure distinction betweenindividuals extracted from different databases. Thus, the URIof the individual ci contains the name of the database and thevalues of the primary key attributes in the table resulting fromffln

x Tx. Eq. (1) describes the procedure to match class individuals,and Eq. (2) shows how to build the URI for a specific instance,where k1, k2, . . ., kn are the primary key attributes of the tables,and r and � stand for the selection and projection operators,respectively.

M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189 4185

ci ¼ frHC ðfflnx TxÞgi ð1Þ

URIðciÞ ¼ D� pk1 ;k2 ;...;knðrHC ðfflnx TxÞÞ

� �i ð2Þ

� The data properties for an individual ci are obtained as a concat-enation (cartesian product) of the values of one or more attri-butes in the row corresponding to ci in Eq. (1). Eq. (3) showsthe matching for a selected data property P with a set of attri-butes {al} for individuals of class C, where c

kj

i if the value ofthe primary key attribute kj for individual ci. The set of attri-butes {al} for the matching are provided by an expert.

PðciÞ ¼ pfalg rHC^k1¼c

k1i^k2¼c

k2i^���^kn¼ckn

i

ðfflnx TxÞ

� �ð3Þ

� An object property O(ci) returns the URI of a set of individuals{bj} of class B that are related with ci by means of such property.Let us consider the matching bj ¼ rHB ðfflm

y SyÞn o

jfrom Eq. (1).

Then the individuals bj resulting from O(ci) are computed as sta-ted in Eq. (4), where the set of tables {Rq} are additional tablesprovided by an expert to perform the join between individualsof classes C and B, and HR are constraints over attributes infflz

qRq.

OðciÞ ¼ D

� pk1 ;k2 ;...;km rHB^HR^HC ððfflmy SyÞÞ ffl ðfflz

qRzÞ ffl ðfflnx TxÞ

� �ð4Þ

Algorithm 1: Procedure for database and ontology mapping

for all database d 2 {D} dofor all ontology class C matched in the ontology do

Compute individuals fcig ¼ rHdCðT�dÞ

n o

for all ci in rHdCðT�dÞ

n odo

Compute the identification id(ci)Add class instance ci for C in the ontologyfor all data property P of class C do

Compute property value P(ci)Add data property value to ci in the ontology

end forend for

end forfor all ontology class C matched in the ontology do

for all class instance ci in C dofor all object property O of class C do

Compute property values fojig ¼ OðciÞ

for all object property instance oji in foj

ig do

Add object property value oji to ci in the ontology

end forend for

end forend for

end for

Once the matching is done, the procedure in Algorithm 1 de-scribes the mapping from the database information to ontology in-stances. For each database matched, all class individuals aremapped firstly in order to avoid inconsistency problems in objectproperties. After that, data and object properties are added to indi-viduals within the ontology. For further reading about the match-ing method and the hierarchy preservation during the matchingwe suggest reading reference Cuéllar et al. (accepted forpublication).

3. Making social network analysis in e-learning

In this section, we assume that the LMSs databases and theontology have been matched and also the data have beenmapped into the ontology instances. Then steps 1–4 in Fig. 2have already finished. In order to fulfill social network analysis,the user of OntoLMS must choose some object properties and/or classes of interest. The hierarchical structure and inheritanceof classes and properties in the ontology plays here an importantrole that allows us to obtain different social networks dependingon the elements selected for the study. Moreover it also eases tomake a fine grained or high level analysis. We illustrate thisidea with some examples based on the ontology structure inFig. 3:

� If the user chooses to study relation foafLMS:TeachesInSub-ject, he will perform a fine grained analysis over teachersthat teach in subjects since the relation and theclass instances that fulfill the requirements are verylimited.� In the opposite situation, let us suppose that the user chooses

to study relation foafLMS:knows, whose range and domain isfoaf:Agent. This could be considered to make a high level anal-ysis, since the selection of such property encompasses a widevariety of classes that are agents (groups, students, departments,subjects, etc.) and also a wide range of properties which are theproperty knows (StudiesSubject, IsWithDepartment, IsFriendOf,etc.)� In an intermediate case, we could assume that user chooses

to study relation foafLMS:knows, classes foafLMS:Teacher andfoafLMS:Student as property range, and class foafLMS:Studentas property domain. The resulting social network wouldinclude nodes which may be either teachers or students andall subproperty relations such as IsStudentOfTeacher, IsFrien-dOf, Teaches, to show any way in which a teacher or a studentcould knows any other student.

We would like to empathize that if two or more relations arechosen for the study, then the social network obtained ismultiplex and the resulting network from OntoLMS is amultigraph.

3.1. Obtaining the social network

If we assume that the user has selected the classes and/orproperties to be studied, the social network is extracted fromthe ontology automatically. During the extraction, each class in-stance is modelled as a node of the graph that represents thenetwork. Each node is labelled with the URI of the individualit matches in the ontology, and its data properties. The objectproperties are the edges that tie two nodes, and the label ofan edge is the name of the property. With these assumptions,the extracted network could be represented as a 3-D N �M � Pmatrix where N is the total account of ontology individuals inthe domain of object properties, M is the number of individualsin the range, and P the number of properties of interest.Algorithm 2 describes the procedure to build the matrix,where {UD}, {UR} and {UP} are inputs for the algorithm and standfor the sets of domain, range classes, and object properties as-serted by the user, respectively. The terms dom(x) and ran(x)are functions that return the domain and range classes ofproperty x. Finally, ind(x) provides the ontology individuals forclass x.

4186 M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189

Algorithm 2: Procedure to extract the social network

if {UD} – ; then

set CD ¼ fUDg \ ð[idomðUPi ÞÞ

else

set CD ¼ [idomðUPi Þ

end ifif {UR} – ; then

set CR ¼ fURg \ ð[iranðUPi ÞÞ

else

set CR ¼ [iranðUPi Þ

end if

set N ¼ jindð[iCDi Þj;M ¼ jindð[iC

Ri Þj; P ¼ jU

P jset l = create matrix of size N �M � P and initializes lijk = 0for all Properties UP

k 2 UP do

for all Individuals i 2 indðCD \ domðUPkÞÞ do

for all Individuals j 2 UðkPðiÞÞ doset lijk = 1

end forend for

end forreturn l

The lines 1–10 in Algorithm 2 compute the set of classes whoseindividuals should appear in the social network: CR are propertyrange classes and CD are domains. After that, the lines 11–12 createthe matrix l that encodes the network. Finally, the lines 13–19walk over the ontology to find class individuals that match theproperties to be studied. If two individuals i and j have a relationk then lijk = 1; otherwise lijk = 0. We would like to remark thatthe graph obtained is a directed graph since the object propertiesmay not be reflexive. On the other hand, there may be situationsin which the graph obtained is undirected, and the same 3-D ma-trix representation is still valid. In this last case each submatrix lijk,1 6 k 6 P, is symmetric.

4. Aspects of interest in social network analysis for e-learning

Undoubtedly, there are many aspects that someone could con-sider interesting to study in an e-learning ecosystem. In this sec-tion, we review some classic analyses carried out over socialnetworks, and give some interpretations of basic and well knowngraph properties for Learning Management Systems using exam-ples. In our examples we illustrate some tools that SNA could offerin e-learning, but also that the interpretation of the results is sub-jective in many cases and the analyst should be suitably trained toobtain real and correct conclusions.

4.1. Degree of nodes

The degree of a node in an undirected graph is the number of di-rect ties that the node has. In the particular case of directed graphs,it may be important to distinguish between in-degree and out-de-gree. A node with a high degree might mean that such a node has aspecial relevance in the network. Let us translate the meaning ofdegree to an e-learning ecosystem involving teachers, subjects,learners and learning materials, to show some examples:

� Considering relation IsFriendOf between students, a high degreein a node could mean that the person it represents has a lot offriends. The analyst might conclude that this student has highsocial abilities.

� Considering relation Teaches between teachers and students, ahigh out-degree in a teacher would suggest that his teachingwork load is high since he has many students. On the otherhand, a high in-degree in a student might mean that the studenthas a lot of teachers, maybe due to either he is assigned withmany subjects or his subjects are usually shared by more thanone teacher.� Considering relation IsInterestedIn between people and learning

material, a learning material node with a high degree suggeststhat it is very popular between teachers and students. Forexample, it could be a basic manual of some general knowledge,a popular note shared by different subjects, etc.

4.2. Distance, walks, paths and reachability

The graph property distance between two nodes measures howfar these nodes are. The interpretation of distance depends on itsown definition and its usefulness over the social network. The eas-iest formulation of distance over graphs is the number of edgesthat tie two known nodes over a walk. A walk is the set of edgesand vertices that join those two nodes. A path is a particular typeof walk where each node and edge in the network is used at mostonce. Some important issues of study are the minimum distancebetween nodes and shortest paths. Finally, a node is reachable fromanother one if there is a path connecting each other.

In e-learning, distance and walks could be used to achieve mul-tiple purposes, specially combined with constraints over the walksand other features like clustering. For example:

� Let us consider teachers, learning material and issues as nodes ina network, where edges model relations made between teachersand learning material, IsAbout between a learning material andan issue, and IsSimilarTo between issues. An analysis over thisnetwork might put in manifest teachers that share topics intheir subjects, maybe with the purpose of finding new col-leagues, to improve their learning materials together, to createa work net, etc.� The same example is also valid for students. In this case, the

learner could find people with the same interest with the pur-pose of sharing notes, exercises, etc.

4.3. Eccentricity and diameter

The eccentricity of a node is the greatest number of edges in theshortest path between this node and the others. The diameter of anetwork is defined as the maximum eccentricity. As an example,the eccentricity might be used in an e-learning system to studyhomogeneity in people that belong to group works, to find infor-mation transference between teachers or students, to find dissim-ilarities between learning resources, etc. Moreover, social networkswith a small diameter are supposed to have high homogeneity be-tween the nodes they have, and also a high connectivity and com-munication between nodes.

4.4. k-Connectivity

The term connectivity attempts to measure reachability be-tween nodes. It is said that a network is k-connected if every pairof nodes in the network are still connected by a path after remov-ing any set of k edges. For instance, connectivity could be used tofind critical nodes in the system: students with low participationin group tasks, unpopular learning materials or issues, etc.

4.5. Centrality

The centrality of a node provides the relevance of the node in thesocial structure. In SNA, central nodes are usually key people that

M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189 4187

join different groups of people, a measure of popularity, etc. Forany learning environment, the study of centrality could offer re-sults about hot issues or topics, design of managers for studentwork groups or subjects, finding popular learning resources, etc.

4.6. Partitioning and clustering

A partition is a classification of the nodes in a network, accord-ing to any known criteria. Each node is assigned to a set thatencompass all actors that share a common property. In an e-learn-ing system, clustering techniques could be useful to group students

Fig. 4. Architecture for ontology match

Fig. 5. Snapshot of the object

according to a shared interest or feature, in order to find peopleprofiles, to find similar learning resources to bring recommenda-tions for students or teachers, etc. In general terms, to find similar-ities and dissimilarities between actors for a given partitionalprocedure.

The properties of above are a small example of popular tools inclassic social network analysis and graph theory, but the possibil-ities go further. If the reader is interested in SNA techniques, wesuggest reading references de Nooy et al. (2004), Hanneman(2005), Huisman and van Duijn (2005), Izquierdo and Hanneman(2006).

ing and social network extraction.

property matching frame.

Fig. 6. Snapshot of the social network settings frame.

4188 M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189

5. Example and implementation remarks

Our goal in this section is to show the application of our ap-proach over a small case study. The example has been generatedsynthetically in order to improve the reading and understanding,since the number of actors in a real LMS would be potentially hugeand difficult to analyze and illustrate in a few pages. The settingsfor our experiments are the following:

� Regarding the individuals for class Teacher, 130 instances wereinserted in the database and matched with the ontology. Foreach teacher, the label for the graph node is the surname dataproperty, named with TXX, where XX stands for the number ofteacher (1–130).� For class Student, 600 instances were generated for this class in

the same way than class Teacher. The label for the graph node ofa student is also the surname data property, named with STXX,where XX stands for the number of student (1–600).� The number of subjects is 45, named with SXX (data property

name).� The number of departments is 4, named with DXX (data prop-

erty name).

In our experiments, each teacher was allowed to be member ofan only department randomly assigned. In addition, each subjectmay contain 60 students and a teacher as much, and there is a sin-gle work group for a subject. Both teachers and students are ran-domly signed up into a subject. Finally, it is allowed that adepartment may offer more than one subject.

The architecture of OntoLMS is illustrated in Fig. 4. Firstly, theontology and the e-learning systems databases are matched, andthe data are mapped into the ontology. Modules LMS Settings Edi-tor, Ontology Loader, Matching Editor, and Mapping Procedure fulfillthese tasks (Section 2.2). The matching editor gives support to makeclasses, data properties and object properties separately in differ-ent frames, using the ontology matching module. As an example,Fig. 5 shows the object property matching appearance for the prop-erty teachesInSubject between teachers and subjects: the user mustchoose the property to match, and the domain and range classes inthe matching. After that, the tables for the matching are selected.For debugging and tracking reasons, the window also shows the ta-

bles matched with the domain and range classes, and the SQLquery that will be generated to do the mapping from the databaseinto the ontology.

Once the data have been saved into ontology instances, the useris allowed to start the Social Network Settings Editor (Fig. 6). Here,the properties of study for the social network are selected withthe domain and range classes of interest. It is also allowed tochoose data properties of the classes selected in order to labelthe actors in the social network, and to restrict the number of ac-tors to appear to a subset of class individuals, using the buttonsnamed Edit in the domain and range classes selection lists.

Finally, the button Extract Social Network starts the Social Net-work extraction method explained in Section 3.1. The result is aplain text file containing the adjacency matrix of the network. Asan example, Eq. (5) shows the matrix obtained for the social net-work analysis over property isWithDepartment, between teachersand departments. In this matrix, a column is assigned with adepartment, and a row with a teacher. We have chosen this outputfile format in order to ease the loading from most mathematicalgeneral software like Mathematica, Matlab or Excel.

D1 D2 D3 D4T1 1 0 0 0T2 0 1 0 0T3 0 1 0 0T4 0 0 1 0T5 1 0 0 0� � � � � � � � � � � � � � �T130 1 0 0 0

ð5Þ

6. Conclusions and further work

Social network analysis is today a hot topic across the web.Most social networks theory methods have been developed dec-ades ago in other disciplines, but the advances in Semantic Weband user experiences in Internet have attracted interest from re-search areas in computer science in the last years. The easy repre-sentation of a network, efficient methods for analysis and the hugeamount of data existing in the web is a challenge for research

M.P. Cuéllar et al. / Expert Systems with Applications 38 (2011) 4181–4189 4189

regarding data mining, knowledge discovery, business models test-ing, psychology or knowledge engineering between others. How-ever, a gap is found when we try to apply SNA over the existingdistributed data storage models. For example, relational databasesintegration and their semantic interpretability are also a greatproblem being studied currently.

In this work, we deal with the semantic integration of databasesusing ontologies and its application in e-learning. This integration isonly the start point to obtain social networks from a shared knowl-edge base of databases with different scheme designs. Furthermore,the hierarchical structure of ontologies allows to make SNA at dif-ferent levels, depending on the classes and the relations of interestin our study. We have provided a method to obtain a personalizedsocial network depending on the analyst’s interests in a LearningManagement System users study. Moreover, we have describedsome classic methods of graphs theory and give some suggestionsabout how they could be used to analyze an e-learning ecosystem.

The future of SNA in e-learning is promising, but further workshould be still coming. The implementation of LMSs using SNA fea-tures could provide a great help in virtual teaching environments,specially combined with current pervasive computing techniques.On the other hand, the improvement and dissemination of morepowerful social network representation and analysis techniquesshould be tested in real applications. For example, fuzzy social net-works would overcome the crisp limitations of classic networks,and the use of hypergraphs could give a major expressivity inapplications in which actors have third-party relations. Moreover,the mathematical category theory could fill some gaps about thereasoning over the network, its representation and its interpreta-tion. We think that the dynamic evolution of social networks inthe Internet is also a promising issue. Methods for dynamic socialnetwork analysis already exist, and their application over e-learn-ing systems in particular and in general the WWW could be usefulto study virtual communities evolution.

References

Bhatia, M., Gaur, P. (2008). Statistical approach for community mining in socialnetworks. In IEEE international conference on service operations and logistics, andinformatics (Vol. 1, pp. 207–211).

Breslin, J., & Decker, S. (2007). The future of social networks on the internet: Theneed for semantics. IEEE Internet Computing, 11(6), 86–90.

Brickley, D., & Miller, L. (2007). FOAF Specification. <http://xmlns.com/foaf/spec/>.Cantador, I., & Castells, P. (2006). Multilayered semantic social network modeling by

ontology-based user profiles clustering: Application to collaborative filtering. InS. Staab & V. Svtek (Eds.), EKAW. Lecture notes in computer science (Vol. 4248,pp. 334–349). Springer.

Cuéllar, M., Delgado, M., & Pegalajar, M. (accepted for publication). A commonframework for information sharing in e-learning management systems. ExpertSystems With Applications.

de Nooy, W., Mrvar, A., & Batagelj, V. (2004). Exploratory social network analysis withPajek. New York, NY, USA: Cambridge University Press.

Dietze, S., Gugliotta, A., & Domingue, J. (2007). Towards adaptive e-learningapplications based on semantic web services. In Proceedings of workshopTENCompetence open workshop on service oriented approaches and lifelongcompetence development infrastructures, Manchester, UK (pp. 75–83).

Dunkel, J., Bruns, R., & Ossowski, S. (2006). Semantic e-learning agents supporting e-learning by semantic web and agent technologies. Springer-Verlag. pp. 237–244.

Dzbor, M., Stutt, A., Motta, E., & Collins, T. (2007). Representations for semanticlearning webs: Semantic web technology in learning support. Journal ofComputer Assisted Learning, 23(1), 69–82.

Ehrlich, K., & Carboni, I. (2005). Inside social network analysis. Copyright 2005, IBM.All rights reserved. 05-10, IBM, Amherst, MA.

Ganley, D., & Lampe, C. (2009). The ties that bind: Social network principles inonline communities. Decision Support Systems, 47(3), 266–274.

Hamasaki, M., Matsuo, Y., Nishimura, T., & Takeda, H. (2007). Ontology extractionusing social network. In International workshop on semantic web for collaborativeknowledge acquisition. p. poster.

Hanneman, R. A. (Ed.). (1988). Computer-assisted theory building: Modeling dynamicsocial systems. Thousand Oaks, CA, USA: Sage Publications, Inc.

Hanneman, R. A. (2005). Introduction to social network methods. <http://faculty.ucr.edu/hanneman/> [Published in digital form].

Hensen, D., & Neville, J. (2003). Data mining in social networks. In Proceedings ofworkshop on dynamic social network modeling and analysis (pp. 289–302).Washington, DC, USA: The National Academies Press.

Henze, N. (2005a). E-learning in the semantic web: The personal learning objectreaders. In Proceedings of second joint workshop on cognition and learning throughmedia-communication for advanced e-learning 2005, Tokyo, Japan, 28th–30thSeptember 2005. Japanese Society for Information and Systems in Education.<http://rewerse.net/publications/rewerse-publications.html#REWERSE-RP-2005-109>.

Henze, N. (2005b). Personalization services for e-learning in the semantic web. InProceedings of second international workshop on adaptive systems for web-basededucation: Tools and reusability, Amsterdam, The Netherlands, 18th–22nd July2005.

Huang, W., Webster, D., Wood, D., & Ishaya, T. (2006). An intelligent semantic e-learning framework using context-aware semantic web technologies. BritishJournal of Educational Technology, 37(3), 351–373.

Huisman, M., & van Duijn, M. (2005). Software for social network analysis. In P.Carrington, J. Scott, & S. Wasserman (Eds.), Models and methods in social networkanalysis. New York: Cambridge University Press.

Izquierdo, L. R., & Hanneman, R. A. (2006). Introduction to the formal analysis ofsocial networks using mathematica. <http://www.luis.izquierdo.name>[Published in digital form].

Jovanovic, J., Gasevic, D., Brooks, C., Devedzic, V., Hatala, M., Eap, T., et al. (2007).Using semantic web technologies to analyze learning content. IEEE InternetComputing, 11(5), 45–53.

Jung, J. J., & Euzenat, J. (2007). Towards semantic social networks. In ESWC’07:Proceedings of the fourth European conference on the semantic web (pp. 267–280).Berlin, Heidelberg: Springer-Verlag.

Jung, J. J., Koo, C.-M., & Jo, G.-S. (2007). A divide-and-conquer approach to detectinglatent community of practice from virtual organizations. In ICCIT’07: Proceedingsof the 2007 international conference on convergence information technology(pp. 129–134). Washington, DC, USA: IEEE Computer Society.

Kanawati, R., & Malek, M. (2007). Computing social networks for informationsharing: A case-based approach. In HCI 15. In D. Schuler (Ed.). Lecture notes incomputer science (Vol. 4564, pp. 86–95). Springer.

Makrehchi, M., & Kamel, M. S. (2006). Learning social networks from webdocuments using support vector classifiers. In WI’06: Proceedings of the 2006IEEE/WIC/ACM international conference on web intelligence (pp. 88–94).Washington, DC, USA: IEEE Computer Society.

Makrehchi, M., & Kamel, M. (2007). Learning social networks using multipleresampling method, 406–411.

Martn, M. S., & Gutierrez, C. (2009). Representing, querying and transforming socialnetworks with rdf/sparql. In European semantic web conference. Lecture Notes inComputer Science (Vol. 5554, pp. 293–307). Springer.

Mayer, A. (2009). Online social networks in economics. Decision Support Systems,47(3), 169–184.

Mitchell-Wong, J., Kowalczyk, R., Roshelova, A., Joy, B., & Tsai, H. (2007). Opensocial:From social networks to social ecosystem. In Digital ecosystems and technologiesconference.

Newman, M. E. J. (2003). The structure and function of complex networks. SIAMReview, 45, 58.

Nilsson, M., Palmr, M., & Naeve, A. (2002). Semantic web metadata for e-learning -some architectural guidelines. In 11th conference on the world wide web(WWW’02), Honolulu (pp. 7–11).

Park, H. (2003). Hyperlink network analysis: A new method for the study of socialstructure on the web. Connections, 25(1), 49–61.

Ross, C., Orr, E. S., Sisic, M., Arseneault, J. M., Simmering, M. G., & Orr, R. R. (2009).Personality and motivations associated with facebook use. Computers in HumanBehaviour, 25(2), 578–586.

Staab, S., Domingos, P., Mika, P., Golbeck, J., Ding, L., Finin, T., et al. (2005). Socialnetworks applied. IEEE Intelligent Systems, 20(1), 80–93.

Thushar, A., & Thilagam, P. (2008). An rdf approach for discovering the relevantsemantic associations in a social network. In 16th international conference onadvanced computing and communications (pp. 214–220).

Torniai, C., Jovanovic, J., Gasevic, D., Bateman, S., & Hatala, M. (2008). E-learningmeets the social semantic web. In Proceedings of the 2008 eighth IEEEInternational conference on advanced learning technologies (pp. 389–393).Washington, DC, USA: IEEE Computer Society.

Weaver, A. C., & Morrison, B. B. (2008). Social networking. IEEE Computer, 41(2),97–100.

Yli-Luoma, V., & Naeve, A. P. (2006). Towards a semantic e-learning theory by usinga modelling approach. British Journal of Educational Technology, 37(3), 445–459.

Zhdanova, A. V., Predoiu, L., Pellegrini, T., & Fensel, D. (2007). A social networkingmodel of a web community. In Proceedings of the 10th International Symposiumon Social Communication (pp. 537–541).

Zhou, C., Chen, H., & Yu, T. (2008). Social network mashup: Ontology-based socialnetwork integration for statistic learning. In IEEE international conference oninformation reuse and integration (pp. 143–146).