Link Prediction for Course Recommendation Integrating Course Relevance and Learning Network Structure Information

ABSTRACT

As learning resources increase exponentially on the internet, it is difficult for learners to find appropriate resources in an online learning platform. As one of the common recommending paradigms, link‐prediction‐oriented recommendation can help learners quickly locate the courses they want to learn, which also helps to enhance learners' learning purpose and learning efficiency. To improve the recommendation quality and reduce the dropout rate of online learning, this study proposes a link‐prediction‐based course recommendation method (LPCR). In LPCR, a heterogeneous learning network is constructed according to the learning relationships among learners and courses. Candidate courses are first explored via multi‐step greedy walk guided purely by course relevance. After candidate collection, a preference score integrating course relevance and learning network structural information is computed for candidate courses to generate recommendations. In this study, a series of experiments have been carried out. Comparative experimental results indicate that the proposed method can produce reasonable course suggestions for target learners and achieves observable improvement in average precision compared with other baseline methods.