Personal Knowledge Graph-enhanced Course Recommendation

With the rapid growth of online learning systems, the demand for personalized course recommendations has been rising steadily. Learner modeling plays a key role in meeting this need. This paper introducing a Personal Knowledge Graph (PKG)-enhanced Course Recommendation System (PKGCR) that improves recommendation accuracy by effectively integrating multimodal information in online education. The system combines ERNIE for text encoding and LightGCN for graph structure learning to capture both textual and structural features. We construct a PKG with five types of relationship to enable fine-grained modeling of user-course interactions and design a dual-path feature learning framework for extracting textual semantic and graph structures. Additionally, we implement an adaptive feature fusion mechanism to dynamically integrate feature from different sources. Experiments on the XuetangX show that PKGCR outperforms state-of-the-art (SOTA) methods in three key metrics: HR@5, NDCG@5, and MRR, demonstrating its effectiveness.