Learning-Motivation-Boosted Explainable Temporal Point Process Model for Course Recommendation

Abstract

Course recommendation is vital for improving students' learning efficiency. In the learning process, students' interests evolve, learning cycles and course scheduling are closely related to temporal information. However, previous course recommendation methods discard it as irrelevant, leading to poor recommendation performance. In addition, the lack of explainability of the course recommendations reduces students' engagement and trust in online learning. To solve two problems, this paper proposes a Learning-motivation-boosted Explainable Temporal point process model for Course Recommendation (LETCR). Firstly, LETCR considers the timestamps in interaction records as absolute time and the sequence of records as relative time, and it calculates the different contributions of historical interaction records to the recommendation results. Secondly, LETCR proposes four factors that affect students' course selection from the perspective of learning motivation: interest preference, follow relationship, conformity and popular course. Finally, LETCR models these with a temporal point process, so as to improve model's explainability. Extensive experiments on the MOOCCourse dataset show that LETCR outperforms other advanced recommendation models by 7.09% and 9.28% on R@10 and NDCG@5, respectively, and has high explainability.