Fuzziness and Semantic Web Technologies

in Personalized eLearning

Marek Reformat

1and Patricia Boechler

2

1Electrical and Computer Engineering, University of Alberta, Edmonton, Canada T6G 2V42Educational Psychology, University of Alberta, Edmonton, Canada T6G 2V4

Abstract

The rapid development of information technologieshas created suitable environment for constructionof eLearning systems. The pivotal point of thosesystems lies in web technologies. Due to thosetechnologies eLearning leads to a number of im-provements in: delivery, access, distributed author-ity, and personalization, when compared with tra-ditional training. Additionally, integration of fuzzi-ness with processes of customization and selectionof adequate material for the user creates a chanceto built truly personalized and adaptive systems.

The paper presents an overview of an architectureof eLearning and its basic component – eLerningEngine – taking a full advantage of ontology, tag-ging, and users’ feedback represented with linguisticdescriptors and quantifiers.

Keywords: eLearning, CI, Sematic Web

1. Introduction

The concept of Semantic Web proposed by Berners-Lee [1] introduces an idea of machine-processableinformation – ontology [7]. This form of representa-tion enables better and more semantic-oriented pro-cessing of information, as well as reasoning aboutit. Its application to eLearning creates opportuni-ties to built systems capable of analyzing learners’needs and behaviors, and more accurate selectionof learning materials. However, in spite of thoseabilities, there is a need to deal with missing orinaccurate data [30]. Learners themselves bring un-certainty – they are imprecise in expressing theirneeds and opinions. Their decisions regarding selec-tion of most suitable alternatives heavily depend oncurrent circumstances, their understanding of situ-ations, and their needs and requirements – thingsthat are “equipped” with ambiguity.

In many situations, identification of the mostsuitable alternative is influenced by opinions pro-vided by others. In such circumstances, a selec-tion process is a multi-criteria decision-making pro-cess where each alternative is evaluated based onmultiple measures originated from multiple sources.Learners are interested in other’s experiences – so-cial software becomes an important vehicle for com-munication and exchange of information among in-

dividuals. Its integration with eLearning seems nat-ural and highly beneficial.

The paper is a description of a framework, and itsdevelopment phases, for constructing human-centriceLearning systems – systems with capabilities torecognize learners’ learning styles and adapt tolearners’ needs and preferences. Such systems com-bine 1) technologies of the Semantic Web – ontologyand forms of its representation, 2) aspects of so-cial software – blogs and tagging, and 3) techniquesof Computational Intelligence (CI) – fuzziness andmulti-criteria decision-making. Additionally, suchsystems provide instructors with abilities to entertheir suggestions and recommendations, and ob-serve learners’ learning activities and comment onthem. It is apparent that a proper representationof data and adequate processing of information arenecessary steps leading towards knowledge-orientedsystems.

The ability to find and represent di�erent typesof relations between pieces of information is a nec-essary condition for creating semantically consciousapplications. This semantic awareness allows formore accurate identification of relevant information.At the same time building any type of system thatinteracts with a human requires ability to handleimprecision and ambiguity. From this point of viewthe application of fuzziness and approximate rea-soning creates a promising avenue of introducing hu-man aspects to software systems, and developmentof more human-aware and human-like systems.

2. Related work

Development of eLearning systems and supportingthem technologies represents an ongoing challengeof fundamental interest and practical relevance. Ex-isting approaches in this area are quite diversifiedenjoying the reliance on various methodologies ande�ective algorithmic developments. A substantialnumber of them adhere to the fundamentals of gen-eral schemes of web technologies.

The domain of eLearning is progressing rapidly.The Semantic Web proposed knowledge representa-tions – RDF (Resource Description Framework) [32]and ontology – quickly finds their way in eLearn-ing applications. Such aspects as formal taxonomiesexpressed with web ontology languages RDFS andOWL [33], and rules represented using the web rule

8th Conference of the European Society for Fuzzy Logic and Technology (EUSFLAT 2013)

© 2013. The authors - Published by Atlantis Press 818

language RuleML [34] play a key role in enablingthe representation and the dynamic construction ofshared and re-usable learning content [25], [23], [12].Some of the highly visible trends are worth brieflyrecalling here. An idea of ontology-enabled anno-tation becomes an important element of eLearningsystems. A number of ontologies representing dif-ferent aspects of knowledge are used to support per-sonalized annotation, real-time discussion, and se-mantic content retrieval [28]. Specialized annota-tion tools are developed [4]. Successful e-Learningsystems have to address how individuals may wantto learn. So, e-Learning designs need to supportpersonalization [15]. Personalized hypertext rela-tions powered by reasoning mechanisms over dis-tributed RDF annotations have been proposed in[9]. Several ontologies have been used which cor-respond to the components of an adaptive system:a domain, a user ontology, and an observation on-tology. In [5] a student model has been developedand integrated with an ontology, enabling the per-sonalization system to guide the student’s learningprocess. Ontology-based solutions for dynamic as-sembly and personalization of learning content partshave been also proposed in [11]. An integratedlearning environment called TANGRAM has beendeveloped. It relies on two ontologies for represent-ing learning object content structure and contenttype (i.e. pedagogical role). A user model ontologyis defined to represent relevant information aboutTANGRAM’s users. Applications of Web Intelli-gence and Artificial Intelligence in eLearning sys-tems enhance adaptivity and enhanced learner com-fort. They enable course sequencing and materialpresentation, automatic discovery, invocation, andcomposition of educational Web services [2].

Social networks make a great basis for developingnew generation of education portals. They providea structure for exchange of information, and com-munication between authors, teachers, and educa-tional institutions [18] [31]. Such networks allowfor combining educational portals, ontologies, andsearch agents with functions such as Web mining,and knowledge management to create, discover, an-alyze, and manage the knowledge of di�erent do-mains presented in educational material.

Multiple aspects of eLearning has been addressedby techniques involving fuzziness. Fuzzy-basedmethods are used for user profiling, determininglearners profiles, evaluating quality of eLearningsystems, as well as enhancing their capabilities. In[16], the authors use fuzzy terms to describe peda-gogical resources as well as users’ profiles. They pro-pose application of fuzzy operators to identify linksbetween available resources and users’ preferences.It leads to flexibility and customization of learningprocesses. In [10], the author applies fuzzy clus-tering to determine categories of learners’ profiles.He performs mining of Web log files to categorizelearners’ behavioral patterns. Another interesting

application of fuzziness can be found in processes ofevaluating quality of eLearning systems [29]. Themethod relies on analysis of questionnaires com-pleted by learners. Fuzzy relations and weightedfactor are used to determine learners’ perceptionsof quality of learning experience.

3. Concept

The ultimate objective of this work is to developa comprehensive framework for development ofeLearning Engines (eLEngine) required for buildinghuman-centric eLearning Systems. Human-centricaspects will be accomplished via utilization of fuzzi-ness and approximate reasoning that are able to ex-press and process ambiguity and imprecision – twovery characteristic features of human selection anddecision-making activities. These techniques com-bined with RDF and ontology-based representationof knowledge and elements of social networks – blog-ging and tagging – will lead to a new way of design-ing eLearning Systems. Development of such a sys-tem will allow us to draw conclusions of immediatepractical relevance to creation of these systems.

In contrast to the research results reported com-monly in the literature, we are concerned with suchaspects of human nature as imprecision, incompleteinformation, and approximate reasoning in orderto ensure engaging and comfortable, yet practicaland e�cient learning environment. The proposedframework evolves around two observations: 1) om-nipresent ambiguity and imprecision in many formsof information provided and used by humans; 2)multiple sources of information that influence thecontent and the form of course materials, Figure 1:

• personal preferences and feedback – in the formof learner’s goals and his/her learning style, aswell as his/her involvement in content annota-tion, tagging, and contributions to blogs;

• required material, suggestions, and constraintsprovided by instructors;

• feedback, content annotations, notes, and eval-uations contributed by other learners.

These sources of information – quite often equippedwith ambiguity – determine relevant componentsof domain knowledge and influence selection ofthe teaching material that is most suitable to thelearner. This process can be performed in a varietyof ways with di�erent levels of precision and impor-tance assigned to each of these sources. For exam-ple, for courses that are fundamental and need a rig-orous approach, the suggestions and constraints im-posed on the content by instructors will have higherpriority, influence, and precision than preferencesprovided by the learner, as well as the feedbackprovided by other learners. Such an approach of-fers flexibility in the process of constructing coursematerials.

Another important aspect is the ability to cre-ate an environment in which the learner feels that

819

Figure 1: Three sources of information determiningcourse material.

he/she participates at the same time in a one-to-one session with an instructor and has access to thefeedback from his/her peers at the same time. Suchability brings a di�erent and very important dimen-sion to the eLearning experience.

The principal challenge is to combine fuzziness(capable of addressing impressions and approximatereasoning) with the SW technologies (for knowl-edge representation) and with mechanisms of so-cial networking. All those elements are needed toachieve the ultimate goal – to build an environmentfor human-centric learning experience where eachlearner feels comfortable and is keen to learn. Theproposed framework is multifaceted and is based onthe most essential features, such as:

• new ways for multi-domain semantic anno-tation of course material and evaluating thelearning experience that fully exploits socialsoftware techniques;

• novel methods of extracting and representingknowledge based on the integration of CI tech-nologies with new information representationforms, i.e., ontology and RDF triples;

• application of fuzziness and approximate rea-soning to represent, analyze and process learn-ers’ preferences and course evaluations; to iden-tify relevant – to learners’ needs – subjects,courses and materials; and to mimic learners’selection and decision-making processes.

The framework will support and quantify themethodological investigations and ensure correct-ness of developed methods and approaches.

4. Technologies and methodology

4.1. Fuzziness and Semantic Web

Fuzziness provides a unique approach to deal with avery human concept of imprecision. Abilities to usesuch imprecise terms as much, so-so and linguisticquantifiers like most, at least, any make fuzzy-basedmethods most suitable for dealing with human eval-

uation of di�erent items and their description. An-other important aspect of fuzziness is its ability toexpress levels of membership of terms to specificconcepts. Using fuzzy-based mechanisms for pro-cessing and reasoning, we can deduce new facts andtheir levels of belonging to specific categories, aswell as precision levels of their descriptions.

The technologies of the SW provide a new andhighly comprehensive approach to knowledge rep-resentation. The basic framework used to representconcepts, their definitions and instances, as wellas relationships existing between concepts is ontol-ogy. The ontology defined in the context of the SWuses the concept of Resource Description Frame-work (RDF) triple as the foundation of knowledgerepresentation. An RDF triple is a triple subject-predicate-object, where: subject identifies what ob-ject the triple is describing; predicate (attribute) de-fines the piece of data in the object a value is givento; and object is the actual value of the attribute;for example, the triple “John likes books” has Johnas subject, likes as predicate and books as object.A collection of RDF triples can be treated as multi-graph, and as such, an RDF-based data model is avery attractive knowledge representation form.

The increasing involvement of users in providingand evaluating contents of the web is expressed bypopularity of systems that rely on principles of so-cial software. One of the most important aspectsthat leads to such popularity is an easy way of pro-viding and augmenting web content and express-ing individuals’ opinions about it. The two meth-ods that are directly involved in those activities areblogs and tagging. A blog allows its users to postcommentary, descriptions of events, or any othermaterial related to a specific topic or topics. Thosedescriptions could be of any form – text, graphics,or even video. The process of tagging is nothingelse but labeling – annotating – resources [17] [8].It is performed by users who use tags to annotateresources easily and freely without knowing any tax-onomies or ontologies. Tags represent any stringsthat users consider appropriate as descriptions ofresources. Resources, on the other hand, could beany items that have been posted and are accessibleby users. This can lead to an interesting way of de-scribing resourses [19] [27]. Those technologies areapplied to design and develop elements and featuresof eLearning Engines.

4.2. Categorization of learners

Identification of categories of learners is another im-portant aspect of eLearning. Each learner is dif-ferent, and each learns in a di�erent way. Edu-cators recognize this, and express these di�erencesas learning styles, cognitive styles, multiple intelli-gences or cognitive traits. With adaptive learningsystems, two approaches have been used: 1) assess-ing cognitive/learning styles at the outset and pre-senting the system to match the learner’ profile, or

820

2) having no preset initialization of the system andallowing adaptations to occur based on the learners’use of the system only. In order to understand theadvantages and disadvantages of both approaches,both approaches need to be tested.

In regards to the former approach, the exactmethod for characterizing learner profiles is highlydebated. Some learner assessment approaches relyon cognitive style measures arising from psycholog-ical theory [24], that is, measures of general cog-nitive tendencies or approaches that endure acrossnumerous types of stimuli. Others focus on learn-ing styles, categorizations of learners preferences ineducational contexts and finally, some methods em-ploy the measurement of basic cognitive traits (e.g.,working memory capacity) as a means to predictwhat material and style of presentation is desir-able for a particular learner [6]. From a theoreti-cal standpoint, it appears none of these approacheshave been unchallenged in regards to their validityand reliability [24].

Given the inconclusive nature of research in theseareas, we adopt a method that creates coherence be-tween the initially derived learner profiles and themechanisms for updating learner profiles that existwithin this particular adaptive system. We choosea cognitive style measure that reflects the specificmechanisms in our adaptive system and that we be-lieve we can modify it to limit the demands of as-sessment on the learner in the initial phase of usingthe system. For the assessment of the structure ofcontent, we rely on the Wholistic/analytic dimen-sion on the Cognitive Styles Analysis (CSA) [22].This provides stimulus to adaptive mechanisms ofthe system. For the assessment of media prefer-ence, we rely on the Verbal/imagery dimension onthe CSA. This will be linked to the course materialrepresentation mechanisms and the personalizationmechanisms of the system. For social preferences,we will again turn to the Verbal/imagery dimensionof the CSA as research supports its identificationof preference for learning contexts that include so-cial interaction. This measure will become an inputto the individual and collaboration-based selectionmechanisms. The CI technologies and knowledgerepresentation as proposed by SW, and SS createsa unique opportunity to design and develop eLearn-ing systems that are able to “analyze” those mea-sures, and adapt to needs and preferences of eachlearner.

4.3. Representation of course material

Applications of digital technologies in educationmeans that learners experience new ways of interac-tion with their learning environment. The aspect ofe�ectiveness of digital media in a teaching processis of major importance.

The task aims at the exploration of how dig-ital representation of material selectively extendsbut also constrains what a learner sees, experi-

ences and has access to, and how it enhances butalso shapes instructors’ representations and presen-tations of their knowledge in an eLearning system.

Such activities as taking notes, marking impor-tant and/or di�cult parts of presented materials,and writing feedback comments (i.e., confirmation,corrective, explanatory, diagnostic, and elaborativeinformation) are vital components of a learning pro-cess at any level of education. The proposed frame-work will be equipped with a number of techniquesand methods, such as content annotation, blogs,and tagging, to allow users for labelling the teach-ing material and providing their opinions about itscontent. A particular emphasis will be put on non-intrusive ways of inputting information – for exam-ple, via voice.

Information provided by users is stored using theknowledge representation schema based on ontologyand RDF triples. Algorithms are used to processusers’ inputs and to annotate course materials withterms and keywords reflecting users’ opinions andnotes.

4.4. Personalization and contextdependence

One of the essential challenges of eLearning systemsis to satisfy learner’s needs and preferences. It isof critical importance to be able to properly elicitand store information about learners, their likes anddislikes, and what kind of methods, techniques andmedia they enjoy during learning activities. Thetechniques should ensure utilization of two types ofinformation:

• learner’s needs, what he/she already knows,his/her goals, things already done, things leftto do, timing, ability to learn, ways of learn-ing, most suitable media (slides, notes, shortlectures);

• current context – time of a day, an amount oftime a learner can spend, place, ability to lis-ten/watch.

Special mechanisms and techniques are used tokeep track of things that work for the learner andthat he/she likes, his/her comments, and informa-tion about his/her favorite instructors.

All information about an individual learner isstored in a specialized ontology. Such ontology iscreated and maintained for each learner. The mech-anisms that support storing imprecise (fuzzy-based)information are developed. The learner is able toprovide her priorities regarding needs and prefer-ences in a human-like form. Special mechanisms forestimating relevance of that information are pro-posed and validated. We address temporal aspectsof information. Such concepts as “fading out” and“old” are defined and their importance and useful-ness are investigated.

821

4.5. Instructors’ input mechanism

The involvement of instructors in the education pro-cess is irreplaceable. Therefore, the proposed frame-work provides a number of ways that an instructorcan monitor students’ activities and act accordingly.Instructors are able to query the system for itemsrelated to available materials and learners takingtheir courses, as well as read comments provided bylearners and related to the material they prepared.The system also allows instructors to enter answersto learners’ questions, suggest alternative materialto learners, and correct them.

4.6. Individual and collaboration-basedmaterial selection

The process of selecting most suitable lecture ma-terials – choice between multiple versions, multiplesections, multiple media, is the most critical partof any eLEngine-based system. Multiple sources ofinformation about lecture materials have to be eval-uated and ranked based on:

• multiple criteria provided by the learner– learner’s goals and profile (preferences,likes/dislikes), as well as comments and notesgiven to similar course materials;

• instructors’ suggestions, recommendations,and constrains – instructors’ notes, observa-tions and expertise will be used to importantselection criteria;

• multiple evaluations of possible material – theprocess of collecting learners’ feedback via so-cial software methods will lead to an extensiveannotation of material, a schema for integrat-ing annotations and extracting most commonopinions is required; those opinions play therole of criteria during a selection process.

The proposed selection methods mimic human-likeaggregation processes from learners’, other users’,and instructors’ points of view. The levels of im-portance are taken into consideration. The methodshave to deal with imprecision information (evalua-tions, criteria, annotation), di�erent priorities, andconstraints. We examine the following aggregationapproaches: fuzzy-based methods; evidence theory;di�erent aggregation techniques including linguisticbased aggregation. Overall, the proposed selectionmechanisms perform decision-making tasks takinginto account:

• what the instructor thinks is important;• what is important for the learner;• what peers think is important.

5. Description of eLearning system

The architecture of the eLEngine-based system ispresented in Figure 2. In a nutshell, the systemknows what the learner wants (learner’s profile) and

likes (annotations, blogs, tags), knows what instruc-tors suggest and recommend (modifications, sugges-tions, annotations, blogs, tags), and knows whatother learners say about available material (anno-tations, blogs, tags). Based on that knowledge andsyllabuses, the system provides the learner withmost suitable alternatives regarding sets of educa-tion material. The system allows learners to makenotes and record opinions. At the same time in-structors have the ability to monitor the learner andprovide modifications and additions to the materialthe learner is currently using. In order to accom-plish that, all the tasks performed by the systemare divided into three categories:

• multi-domain annotation of course materialstored in a repository combined with tech-niques and methods of extracting importantoptions based on tag clouds, blogs and learners’notes; instructors’ suggestions and constrainsare also used to annotate available material;

• personalization that leads to creation and up-dating of learner’s profile that contains in-formation about learner’s preferences, needs,likes/dislikes;

• prioritization-based multi-criteria selectionthat performs selection of most suitable mate-rial based on learner’s profile, other learners’opinions, and instructors’ inputs.

Figure 2: Architecture of eLEngine-based System.

The annotation activities are presented in Fig-ure 3. All annotation is performed on course ma-terial stored in the repository of the system, and itreflects three domains (dimensions): learners’, in-structors’, and knowledge relevant to course top-ics. The learners-wise annotation is supported bysuch processes as identification of concept and key-words in annotations and blogs, as well as analysisof tag-clouds. The instructor-wise annotation takesinto account all requirements and recommendationsprovided by instructor. The domain-wise annota-tion leads to annotation of all repository materialswith terms originated from specific knowledge do-mains. In the end, all materials is annotated withthree types of terms originated from three sourcesof annotations.

822

Figure 3: Annotation Activities of eLEngine-basedSystem

The personalization activity is also shown in Fig-ure 3. It uses results of the same process as an-notation: concept and keyword identification andanalysis of tag-clouds to extract information that isrelated to a single learner. This information is usedfor updating of the learner’s profile. Both anno-tation and personalization process are continuouslyperformed so annotations and profile is up-to-date.

The multi-criteria selection mechanisms are pre-sented in Figure 4. The first step in the processselects a few sets of alternative materials. The se-lection uses the annotated course material and isbased on learner’s needs and goals, and providedsyllabuses. An important element is extraction ofevaluations of those materials done by other learn-ers. The last and most important step of selectionis identification of the most suitable material. Thisstep relays on a number of di�erent decision-makingmethods that are able to deal with multiple crite-ria with multiple levels of priorities [26], impreciseinformation and human-like aggregation of evalu-ations [21] [20]. These processes can use di�erentsources of information, for example RSS-feeds, anddi�erent analysis techniques including formal ones,for example FCA [3].

6. System development

6.1. Overview

The framework objectives require a structured workplan comprised of a number of parallel and comple-mentary activity lines. Progress within those lineswill be interactive, incremental, and, when neces-sary, iterative. The conceptual activity line domi-nate as the eLearning Engine methodology evolves.It requires a very thorough and extensive explo-ration of the developed concepts that leads to tran-sition into algorithmic development and validation– the algorithmic activity. This is followed by in-tegration and experimentation using existing soft-ware tools – the experimentation activity. For in-stance, validation of software implementing a givenalgorithm provides insights into conceptual founda-

Figure 4: Selection Processes

tions, leading to revision of the algorithm and itsnew implementation. Similarly, improvements inimplementation impact an integration process.

Conceptual activity line: within this activity thefocus is put on the fundamentals and conceptual un-derpinnings, in particular the following aspects areinvestigated: annotation mechanisms, personaliza-tion techniques and a variety of selection schemes,as well as diversity of possibilities in eliciting opin-ions of users – learners and instructors. The mech-anisms of human-centricity, which becomes presentwhen forming a feedback relevance loop, could beconceptually diversified leading to a number of gen-eral pursuits.

Algorithmic activity line: Being cognizant of thenumerous possibilities explored at the conceptuallevel, the algorithmic activity is concerned with theunderlying aspects of their realization and establish-ing their feasibility from the computational stand-point. Likewise, the focus is put on investigationof the realization of the specific conceptual devel-opments (e.g., through a critical assessment of re-quired learning capabilities which are necessary toarrive at the e�ective algorithm). The activitiesfocus on two main groups of tasks performed byeLEngine-based systems: annotation and selection.The annotation tasks are presented in Figure 3. Thecourse materials are constantly annotated based onthree domains: users’ input (annotations, blogs andtagging); instructors’ input (suggestions, modifica-tions, constrains); and knowledge domain (vocabu-laries related specific topics covered by course ma-terials). Closely related is a process of extractinglearner’s profile. It leads to updates of the profilebased on learner’s comments left in blogs, tagging,and notes left during interaction with other materi-als. The selection tasks are shown in Figure 4.

823

Experimentation activity line: This constitutesan important phase of work on the proposed frame-work and is exclusively devoted to the constructionof the software prototype of the eLearning environ-ment which once designed will be exploited as acomprehensive experimentation platform.

6.2. Development phases

The first phase is dedicated to construction of repos-itory for learning materials, development of ontolo-gies, as well as designing and implementation ofmechanisms for concept identification from blogsand tag clouds, and for annotation and selection.All this leads to preliminary prototypes of the sys-tem. Intermittent pilot testing occurs to developand test the reliability of the modified CSA andto inform the design process. A measure of ICTLiteracy will be developed based on [13] and [14]frameworks that include measures of general cogni-tive abilities as well as technical skills.

The second phase focuses on modifications andextensions of already implemented mechanisms, ad-dition of fuzzy-based selection schemas, as wellas development of procedures for identification oflearners’ profiles and evaluation of course materi-als. Based on design changes from the pilot testresults, the first version of the proposed frameworkis tested with a sample of undergraduate educationstudents. At the outset of the study, students com-plete the modified, shortened CSA and the ICT Lit-eracy Test. Students have the opportunity to usethe system on three occasions and will be asked tocomplete specific tasks within the system. After us-ing the system, students complete a questionnaireon the usability of features of the system. Focusgroups with small groups of students (4-6) is con-ducted to provide further data on the student’s ex-periences of using the system (preferences, impres-sions etc.). The full CSA is administered to testfor correlation with the modified version. Students’patterns of use in relation to their measured learnerprofile are examined.

Based on the assessment preformed in the secondphase, the system undergos final design changes –tuning of algorithms and procedures. Eventually,the final version of the system is tested for an ex-tended time period (a four week module in an on-line education course) in order to assess both usabil-ity and learning. Students again complete the modi-fied, shortened CSA and the ICT Literacy Test. Re-lationships between learner profiles, technical skillsand observed user behavior patterns are examined.To ensure that no learner profile is disadvantagedin the final learning assessment phase, learning isassessed with both fact-based multiple choice ques-tions and open-ended conceptually-based questions.Again, the full CSA is administered to test for cor-relation with the modified version. Focus groupsare conducted to access students’ qualitative expe-riences of the system.

7. Conclusion

One of the strengths of the proposed frameworkfor development of eLearning systems is its multi-disciplinary nature. This work leads to interestingresults in several important areas:

• ontology and RDF triples – those are newand conceptually challenging forms of knowl-edge representation; activities related to adap-tation of these forms to eLearning systemsand their integration with interactive systemsshould lead to improvements in system’s abili-ties to store, access and analyze information;

• blogs and tagging – incorporation of these ac-tivities with eLearning systems should improveagility of eLearning systems and their ability toabsorb users’ feedback, additionally their inte-gration with new forms of knowledge represen-tation should lead to a better analysis of infor-mation embedded in blog posts and used tags;

• fuzziness and multi-criteria decision making –as the core technologies of the framework theyshould play an important role in the process ofcreating more human-centric systems; at thesame time the framework should provide anevidence of necessity of application of thesetechniques to development of real-world systemable to support users’ activities in a personal-ized way; the combination of fuzziness with thenew forms of knowledge representation – ontol-ogy and RDF triples – should increase presenceof CI technologies on the Web.

The proposed eLearning framework constitutes avery important step towards direct application offuzziness and new forms of knowledge representa-tion to “real world” needs of eLearning systems.The framework also addresses the challenges im-posed by human-centric systems.

References

[1] T. Berners-Lee, J. Hendler, and O. Lassila,O., The Semantic Web, Scientific American,284:34–43, 2001

[2] V. Devedzic, Web Intelligence and Artificial In-telligence in Education, Educational Technol-ogy and Society, 7 (4):29–39, 2004.

[3] C. De Maio, G. Fenza, M. Gaeta, V. Loia,F. Orciuoli, and S. Senatore. 2012. RSS-basede-learning recommendations exploiting fuzzyFCA for Knowledge Modeling, Appl. Soft Com-put., 12(1):–113-124, 2012

[4] A. Faical, C. Weiqin, and D. Cyrille, Seman-tic Annotation Tools for Learning Material, 4thInt. Conf. on Adaptive Hypermedia and Adap-tive Web-Based Systems, 2004.

[5] P. Gomes, B. Antunes, L. Rodrigues, A. San-tos, and J. Barbeira, Using Ontologies foreLearning Personalization, 3rd ELearning Con-

824

ference - Computer Science Education, Coim-bra, Portugal, 2006.

[6] S. Graf, T. Lin, and Kinshuk, The relationshipbetween learning styles and cognitive traits -Getting additional information for improvingstudent modelling. Computers in Human Be-havior, 24(2):122–137, 2008.

[7] T. Gruber, Toward Principles for the Design ofOntologies used for Knowledge Sharing, Int. J.of H-C Studies, 43(4-5):907–928, 1995.

[8] T. Hammond, T. Hannay, B. Lund, andJ. Scott, Social Bookmarking Tools (I): AGeneral Review, D-Lib Magazine, 11, 2005,http://www.dlib.org/dlib/april05/hammond/04hammond.html

[9] N. Henze, P. Dolog, and W. Nejdl, Reason-ing and Ontologies for Personalized E-Learningin the Semantic Web, Educational Technologyand Society, 7 (4):82–97, 2004.

[10] M.A. Hogo, Evaluation of E-Learners Be-haviour using Di�erent Fuzzy Clustering Mod-els: A Comparative Study, Int. J. of Com-puter Science and Information Security, 7(2):131–140, 2010.

[11] J. Jovanovic, D. Gasevic, and V. Devedzic,TANGRAM for Personalized Learning UsingSemantic Web Technologies, J. of EmergingTechnologies in Web Int., 1(1):6–21, 2009.

[12] J. Jovanovic, D. Gasevic, C. Torniai, S. Bate-man, and M. Hatala, The Social Semantic Webin Intelligent Learning Environments-State ofthe Art and Future Challenges, InteractiveLearning Environments, 17(4):273–308, 2009.

[13] S.S. Kurbanoglu, Self-e�cacy: A con-cept closely linked with information literacyand lifelong learning, J. of Documentation,59(6):635–646, 2003.

[14] L. Markauskaite, Exploring the structure oftrainee teachers’ ICT literacy: the main com-ponents of, and relationships between, gen-eral cognitive and technical capabilities. Educa-tional Technology Research and Development,55(6):, 2007, 547–572.

[15] M. Martinez, High Attrition Rates in e-Learning: Challenges, Predictors, and Solu-tions, The eLearning Developers’ J., 2003.

[16] C. Mencar, C. Castiello, and A.M. Fanelli,Fuzzy User Profiling in e-Learning Contexts, inI. Lovrek, R.J. Howlett, and L.C. Jain (Eds.):KES 2008, Part II, LNAI 5178, 230–237, 2008.

[17] A. Mathes, Folksonomies – Coop-erative Classification and Commu-nication Through Shared Metadata,http://www.adammathes.com/academic/computer-mediated-communication/folksonomies.html

[18] P. Raghavan, Social Networks: From the Webto the identifying some challenging issues anddiscussing Enterprise, IEEE Internet Comput-ing, 6(1):91–94, 2002.

[19] M. Z. Reformat, and R.R. Yager, Tag-basedFuzzy Sets for Criteria Evaluation in On-lineSelection Processes, J. of Ambient Intelligenceand Humanized Computing, 2(1):35–51, 2011.

[20] M. Z. Reformat, R.R. Yager, Z. Li, N. Ala-jlan, Human-inspired Identification of High-level Concepts using OWA and LinguisticQuantifiers, Inter. J. of Computers, Commu-nications and Control, 6(3):473-502, 2011.

[21] M. Z. Reformat, and S. K. Golmohammadi,Rule- and OWA-based Semantic Similarity forUser Profiling, Inter. J. of Fuzzy Systems,12(2):87-102, 2010.

[22] R. Riding, and I. Cheema, Cognitive Styles– An Overview And Integration, EducationalPsychology, 11(3-4):193, 1991.

[23] D.G. Sampson, M.D. Lytras, G. Wagner, andP. Diaz, Ontologies and the Semantic Web forE-learning, Educational Technology and Soci-ety, 7 (4):26–28, 2004.

[24] R.J. Sternberg, and E.L. Grigorenko, A Cap-sule History of Theory and Research on Styles,in Perspectives on Thinking, Learning andCognitive Styles, Robert J. Sternberg and Li-Fang Zhang (Eds), Lawrence Erlbaum Asso-ciates, Mahwah, NJ, Publication 1–22. 2001.

[25] L. Stojanovic, S. Staab, and R. Studer, eLearn-ing based on the Semantic Web, WebNet2001 -World Conference on the WWW and Internet,23–27, 2001.

[26] R.R. Yager, M. Reformat, G. Gumrah, Usinga Web Personal Evaluation Tool - PET forLexicographic Multi-criteria Service Selection,Knowledge-Based Syst., 24(7):929-942, 2011.

[27] R.R. Yager, and M.Z. Reformat, Using FuzzySets to Model Information Provided by SocialTagging, 2010 IEEE World Congress on Com-putational Intelligence, Barcelona, Spain, July18-23, 2010.

[28] S.J.H. Yang, I.Y.L. Chen, and N.W.Y Shao,Ontology Enabled Annotation and KnowledgeManagement for Collabor Learning in VirtualLearning Community, Educational Technologyand Society, 7 (4): 70-81, 2004.

[29] H. Yongqiang, W. Jianxin, A Study on FuzzyEvaluation of E-learning Teaching Quality,2010 2nd Int. Conf. on e-Business and Infor-mation System Security (EBISS), 1–4, 2010.

[30] L.A. Zadeh, Toward a perception-based theoryof probabilistic reasoning with imprecise prob-abilities, J. of Statistical Planning and Infer-ence, 105:233–264, 2002.

[31] N. Zhong, J., Liu, J., and Y.Yao, In Search ofthe Wisdom Web, IEEE Computer, 35 (11):27–31, 2002.

[32] http://www.w3.org/RDF/ (access: March 27,2013)

[33] http://www.w3.org/TR/owl2-overview/ (ac-cess: March 27, 2013)

[34] http://ruleml.org/ (access: March 27, 2013)

825