PreKGAT: Using Course Prerequisite Relation and Knowledge Graph Attention Network for Course Recommendation

Faced with a large number of courses, researchers have been focusing on how to improve the precision of course recommendations. At present, introducing knowledge graphs as auxiliary information into recommendation has achieved remarkable results. However, there is also a prerequisite relationship between courses, and incorporating them into the recommendation system can improve recommendation results. Inspired by graph attention neural networks and course prerequisite relationships, we propose PreKGAT. PreKGAT extracts prerequisite relationships between courses through generated meta-paths, and uses the prerequisite relationships of courses to enrich the course knowledge graph. The entities in the knowledge graph are embedded and propagated to mine the course information related to user preferences. Next, graph convolutional networks are used to aggregate neighborhood features to obtain feature representations of users and courses, and attention mechanisms are used to distinguish the importance of neighborhoods and enhance the embedding representations of nodes. Finally, the ratings between users and courses are predicted. The experimental results showed that compared with the optimal baseline, F1@10 and Mrr@10 increased by 2.26% and 1.08%, respectively.