GADN: GCN-Based Attentive Decay Network for Course Recommendation

Course recommendation in online platforms aims to address the information explosion problem and make personalized recommendations for users. Most of the recent recommendation models mainly strive to model inherent user preference while ignoring the users’ learning process and overlooking the relations among courses (e.g., The prerequisite course for Deep Learning is Linear Algebra). This paper proposes an innovative model named GCN-based Attentive Decay Network for Course Recommendation (GADN). Specifically, (1) we utilize the GCN-based Knowledge Extraction Layer to explicitly model the relationships on Collaborative Sequence Graph (CSG), which incorporates user-item interactions and course sequence information; (2) we incorporate the Knowledge Evolution Layer with a monotonic attention decay mechanism to model users’ learning process. At the Knowledge Evolution Layer, we calculate attention weights using exponential decay and an absolute distance measure, in addition to the similarity between courses; (3) our method has a specific explanation through visualizing attention weight. Systematically, we conduct a series of experiments to demonstrate the effectiveness of our model on several prevalent metrics compared to the other prevailing baseline methods.