A Semantic Enhanced Course Recommender System via Knowledge Graphs for Limited User Information Scenarios

Abstract

The Universal Declaration of Human Rights states that “everyone has the right to education” and it must be “equally accessible to all”. Aligned with these principles are online learning platforms that offer free online courses that require no or minimal information from their students to democratize learning. Despite the benefits, the lack of user information challenges traditional recommendation systems and, when applied to education, impacts student experiences. This paper addresses the problem of limited student information in course recommendation systems. We tackle the problem by generating a semantically enriched vector-based representation of course content using an Open Knowledge Graph (DBpedia) and Topic Modeling methods, such as LDA and LSA and by collecting tracking events from log systems (Google Analytics) to create a hybrid recommendation system using content-based and collaborative filtering strategies for low-information scenarios. Our experiments use GCFGlobal.org, an online learning platform offering free self-paced online courses to 100 million people, to validate our approach. Results indicate that the proposed approach outperforms the previous works in the field contributing to the creation of fairer recommendation systems.