KRec-C2: A Knowledge Graph Enhanced Recommendation with Context Awareness and Contrastive Learning

Knowledge graph (KG) has been widely utilized in recommendation system to its rich semantic information. There are two main challenges in real-world applications: high-quality knowledge graphs and modeling user-item relationships. However, existing methods try to solve the above challenges by adopting unified relational rules and simple node aggregation, which cannot cope with complex structured graph data. In this paper, we propose a Knowledge graph enhanced Recommendation with Context awareness and Contrastive learning (KRec-C2) to overcome the issue. Specifically, we design an category-level contrastive learning module to model underlying node relationships from noisy real-world graph data. Furthermore, we propose a sequential context-based information aggregation module to accurately learn item-level relation features from a knowledge graph. Extensive experiments conducted on three real-world datasets demonstrate the superiority of our KRec-C2 model over existing state-of-the-art methods.