Erdt, Mojisola Helen (2014)
Personalized Recommender Systems for Resource-based Learning - Hybrid Graph-based Recommender Systems for Folksonomies.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung
Kurzbeschreibung (Abstract)
As the Web increasingly pervades our everyday lives, we are faced with an overload of information. We often learn on-the-job without a teacher and without didactically prepared learning resources. We not only learn on our own but also collaboratively on social platforms where we discuss issues, exchange information and share knowledge with others. We actively learn with resources we find on the Web such as videos, blogs, forums or wikis. This form of self-regulated learning is called resource-based learning. An ongoing challenge in technology enhanced learning (TEL) and in particular in resource-based learning, is supporting learners in finding learning resources relevant to their current needs and learning goals. In social tagging systems, users collaboratively attach keywords called tags to resources thereby forming a network-like structure called a folksonomy. Additional semantic information gained for example from activity hierarchies or semantic tags, form an extended folksonomy and provide valuable information about the context of the resources the learner has tagged, the related activities the resources could be relevant for, and the learning task the learner is currently working on. This additional semantic information could be exploited by recommender systems to generate personalized recommendations of learning resources. Thus, the first research goal of this thesis is to develop and evaluate personalized recommender algorithms for a resource-based learning scenario. To this end, the resource-based learning application scenario is analysed, taking an existing learning platform as a concrete example, in order to determine which additional semantic information could be exploited for the recommendation of learning resources. Several new hybrid graph-based recommender approaches are implemented and evaluated. Additional semantic information gained from activities, activity hierarchies, semantic tag types, the semantic relatedness between tags and the context-specific information found in a folksonomy are thereby exploited. The proposed recommender algorithms are evaluated in offline experiments on different datasets representing diverse evaluation scenarios. The evaluation results show that incorporating additional semantic information is advantageous for providing relevant recommendations. The second goal of this thesis is to investigate alternative evaluation approaches for recommender algorithms for resource-based learning. Offline experiments are fast to conduct and easy to repeat, however they face the so called incompleteness problem as datasets are limited to the historical interactions of the users. Thus newly recommended resources, in which the user had not shown an interest in the past, cannot be evaluated. The recommendation of novel and diverse learning resources is however a requirement for TEL and needs to be evaluated. User studies complement offline experiments as the users themselves judge the relevance or novelty of the recommendations. But user studies are expensive to conduct and it is often difficult to recruit a large number of participants. Therefore a gap exists between the fast, easy to repeat offline experiments and the more expensive user studies. Crowdsourcing is an alternative as it offers the advantages of offline experiments, whilst still retaining the advantages of a user-centric evaluation. In this thesis, a crowdsourcing evaluation approach for recommender algorithms for TEL is proposed and a repeated evaluation of one of the proposed recommender algorithms is conducted as a proof-of-concept. The results of both runs of the experiment show that crowdsourcing can be used as an alternative approach to evaluate graph-based recommender algorithms for TEL.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2014 | ||||
Autor(en): | Erdt, Mojisola Helen | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Personalized Recommender Systems for Resource-based Learning - Hybrid Graph-based Recommender Systems for Folksonomies | ||||
Sprache: | Englisch | ||||
Referenten: | Steinmetz, Prof. Dr.-Ing. Ralf ; Lucke, Prof. Dr.-Ing. habil. Ulrike | ||||
Publikationsjahr: | 12 September 2014 | ||||
Ort: | Darmstadt | ||||
Datum der mündlichen Prüfung: | 19 November 2014 | ||||
URL / URN: | http://tuprints.ulb.tu-darmstadt.de/4137 | ||||
Zugehörige Links: | |||||
Kurzbeschreibung (Abstract): | As the Web increasingly pervades our everyday lives, we are faced with an overload of information. We often learn on-the-job without a teacher and without didactically prepared learning resources. We not only learn on our own but also collaboratively on social platforms where we discuss issues, exchange information and share knowledge with others. We actively learn with resources we find on the Web such as videos, blogs, forums or wikis. This form of self-regulated learning is called resource-based learning. An ongoing challenge in technology enhanced learning (TEL) and in particular in resource-based learning, is supporting learners in finding learning resources relevant to their current needs and learning goals. In social tagging systems, users collaboratively attach keywords called tags to resources thereby forming a network-like structure called a folksonomy. Additional semantic information gained for example from activity hierarchies or semantic tags, form an extended folksonomy and provide valuable information about the context of the resources the learner has tagged, the related activities the resources could be relevant for, and the learning task the learner is currently working on. This additional semantic information could be exploited by recommender systems to generate personalized recommendations of learning resources. Thus, the first research goal of this thesis is to develop and evaluate personalized recommender algorithms for a resource-based learning scenario. To this end, the resource-based learning application scenario is analysed, taking an existing learning platform as a concrete example, in order to determine which additional semantic information could be exploited for the recommendation of learning resources. Several new hybrid graph-based recommender approaches are implemented and evaluated. Additional semantic information gained from activities, activity hierarchies, semantic tag types, the semantic relatedness between tags and the context-specific information found in a folksonomy are thereby exploited. The proposed recommender algorithms are evaluated in offline experiments on different datasets representing diverse evaluation scenarios. The evaluation results show that incorporating additional semantic information is advantageous for providing relevant recommendations. The second goal of this thesis is to investigate alternative evaluation approaches for recommender algorithms for resource-based learning. Offline experiments are fast to conduct and easy to repeat, however they face the so called incompleteness problem as datasets are limited to the historical interactions of the users. Thus newly recommended resources, in which the user had not shown an interest in the past, cannot be evaluated. The recommendation of novel and diverse learning resources is however a requirement for TEL and needs to be evaluated. User studies complement offline experiments as the users themselves judge the relevance or novelty of the recommendations. But user studies are expensive to conduct and it is often difficult to recruit a large number of participants. Therefore a gap exists between the fast, easy to repeat offline experiments and the more expensive user studies. Crowdsourcing is an alternative as it offers the advantages of offline experiments, whilst still retaining the advantages of a user-centric evaluation. In this thesis, a crowdsourcing evaluation approach for recommender algorithms for TEL is proposed and a repeated evaluation of one of the proposed recommender algorithms is conducted as a proof-of-concept. The results of both runs of the experiment show that crowdsourcing can be used as an alternative approach to evaluate graph-based recommender algorithms for TEL. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
Freie Schlagworte: | Resource-based Learning, Recommender Systems, Crowdsourcing, Folksonomies, Tagging, User-centric Evaluation, Technology-enhanced Learning, Recommender Algorithms, Information Retrieval, Knowledge Management | ||||
URN: | urn:nbn:de:tuda-tuprints-41374 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik 600 Technik, Medizin, angewandte Wissenschaften > 620 Ingenieurwissenschaften und Maschinenbau |
||||
Fachbereich(e)/-gebiet(e): | 18 Fachbereich Elektrotechnik und Informationstechnik 18 Fachbereich Elektrotechnik und Informationstechnik > Institut für Datentechnik 18 Fachbereich Elektrotechnik und Informationstechnik > Institut für Datentechnik > Multimedia Kommunikation |
||||
Hinterlegungsdatum: | 07 Dez 2014 20:55 | ||||
Letzte Änderung: | 16 Apr 2021 09:52 | ||||
PPN: | |||||
Referenten: | Steinmetz, Prof. Dr.-Ing. Ralf ; Lucke, Prof. Dr.-Ing. habil. Ulrike | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 19 November 2014 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |