A course recommendation method based on bidirectional fusion of graph convolution using knowledge graphs

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Knowledge graph–based recommendation systems, such as KGCN, have been widely adopted to alleviate data sparsity and cold-start issues by incorporating auxiliary semantic information. However, most existing methods model either the user side or the item side independently, failing to fully exploit the complementary information between user preference propagation and course knowledge semantics. To address this limitation, this paper proposes a bidirectional fusion framework, termed KGCN-RN, for course recommendation. The proposed method leverages RippleNet to learn user preference propagation representations over the knowledge graph and integrates them into the user modeling process of KGCN. Meanwhile, semantic information from neighboring course entities is aggregated through knowledge graph convolution, enabling collaborative enhancement of both user and course representations. Experimental results on the public MOOCCube dataset demonstrate that KGCN-RN achieves competitive or superior recommendation performance compared with representative general and course-oriented baseline methods in terms of HR and NDCG. The results indicate that the proposed bidirectional fusion strategy effectively improves recommendation accuracy and ranking quality by jointly modeling learner preference propagation and course semantic aggregation.