Recommender Systems for Social Learning Platforms

Research output: ThesisDissertation (external)


In this thesis, we present our contribution to supporting users of so-called social learning platforms with recommendations. Social learning platforms work in ways similar to social networks such as Facebook; but, unlike Facebook, they are developed exclusively for the purpose of learning and knowledge sharing. Therefore, they often serve as a meeting ground exclusively for educational stakeholders such as teachers, students, learners, policy makers, and so on. Nevertheless, with the increasing amount of data in social learning platforms, the users of such platforms face the information overload problem. Recommender systems can help to deal with this problem. They provide users with suggestions on resources based on their interests and activities. In this research, the main recommendation task is to generate a list of top N resources for each user, based on their social activities within the social learning platforms. Therefore, the main research question in this research is: How to best support users of social learning platforms with resource recommendations based on their interactions within the platform? To answer this research question, we evaluate which recommender system performs best in such platforms with their own data, characteristics, and requirements. description, this research consists of three phases: First,the conceptual model that presents the theoretical aspects and the methodology of the research. Second, the offline evaluation phase in which we investigate which recommender system method can best perform in social learning platforms in terms of the accuracy metrics used in offline studies such as precision, recall, and F-score. And, finally in the third phase, we run an online user evaluation in a real-world setting, with actual users of an authentic social learning platform:Open Discovery Space. In this final phase, we investigate which recommender system can best fit users of social learning platforms in terms of several quality metrics including prediction accuracy of the recommendations. Our findings suggest that user-centric evaluation results are not necessarily in line with data-centric evaluation results. We conclude that the traditional evaluation of recommender systems in terms of prediction accuracyonly does not suffice to judge performance of recommender systems on the user side. Moreover, the user-centric evaluation provides valuable insights in how candidate algorithms perform on each of the five quality metrics for recommendations: usefulness, accuracy, novelty, diversity, and serendipity.
Original languageEnglish
Awarding Institution
  • Open University of the Netherlands
  • Sloep, Peter, Supervisor, External person
  • Drachsler, Hendrik, Supervisor, External person
Award date12 Apr 2019
Print ISBNs978-94-92739-76-6
Publication statusPublished - 12 Apr 2019
Externally publishedYes


Dive into the research topics of 'Recommender Systems for Social Learning Platforms'. Together they form a unique fingerprint.

Cite this