ABiNE-CRS: course recommender system in online education using attributed bipartite network embedding

Personalized course recommender systems in online learning platforms provide courses that fit students’ personal needs using their individual past preferences. Due to their good performance, Collaborative Filtering methods are the most widely used. However, these methods suffer from cold-start, data sparsity and inability to process implicit feedback, which affects the recommendation results. Existing collaborative filtering course recommender systems utilize information from external sources such as contents and high-order relations to solve these challenges. However, they failed to jointly utilize the direct relations between students and courses in the form of implicit feedback, the high- order collaborative relations and the content similarity between them. In this work, we propose a novel method, ABiNE-CRS short for Attributed Bipartite Network Embedding for Course Recommender System. Our model elaborately captures the direct relations between students and courses in the form of implicit feedback, the high-order collaborative and content similarity between a set of students and a set of courses to learn high-quality representations of students and courses for recommendation. Utilizing these relations jointly solves sparsity, cold-start and implicit feedback challenges, thereby improving the overall recommendation result. We conduct experiments to evaluate the performance of ABiNE-CRS with state-of-the-art methods and existing course recommender systems. The results indicate ABiNE-CRS outperforms the state-of-the-art methods with the highest improvement of 2.43% on MRR@10 and 3.35% on RECALL@10. It also outperforms existing course recommender systems with the highest improvement of 2.52% on MRR@10. Our model also shows significant improvement in both student and course cold-start.