Fine-grained relation contrast enhancement of knowledge graph for recommendation

The use of knowledge graphs (KGs) in recommender systems has achieved excellent results. Knowledge graphs, serving as auxiliary information, effectively alleviate issues related to data sparsity and the cold start problem, strengthening the modeling of item sets and the representation of user preferences. However, few works explore the fine-grained implicit relationships within knowledge graphs. In addition, most KG-based recommender systems only jointly model user-item interaction information and knowledge graph information, failing to effectively balance the relationship between these two types of information. To address these problems, we propose a Recommendation Algorithm Based on Fine-grained Relation Contrast Enhancement of Knowledge Graph (RA-FRCE). Specifically, we use the graph neural network to learn representations for nodes in the user-item interaction graph and collaborative knowledge graph. Then, we explore fine-grained implicit relationships in the knowledge graph, mine semantically similar items, and optimize the embedding of nodes through the implicit relationships. Subsequently, we conduct contrastive learning between semantic item information and item information in the collaborative knowledge graph to enhance item information. Finally, we adopt an adaptive adjustment fusion mechanism, dynamically adjusting the weights of interaction graph information and collaborative knowledge graph information to achieve collaborative optimization and adaptive fusion. Extensive experiments on three standard datasets show that our RA-FRCE model outperforms current state-of-the-art baselines. Our implementation codes are available at https://github.com/UPCRS/RAFRCE.