Multi-contrastive Learning Recommendation Combined with Knowledge Graph

Knowledge graphs have rich semantic information and are widely used in recommender systems to alleviate the sparsity problem of user-item interaction data. Existing graph collaborative filtering models based on knowledge graphs face two challenges: i) real-world knowledge graphs usually contain a large amount of noise, which affects the accuracy of recommendation results. ii) the intrinsic association of user-item interaction data is not fully explored, which cannot effectively alleviate the data sparsity problem. To address the above two problems, we proposes a multi-contrastive learning recommendation model combined with knowledge graph (MLKG). First, a knowledge-aware aggregation enhancement method is used to derive a more robust knowledge-aware representation for items, second, the node representation is enhanced by random graph contrastive learning, and finally, the neighbor relationship of nodes is strengthened by neighborhood contrastive learning. Since MLKG makes full use of knowledge semantic information, self-supervised signal and neighborhood structure signal to learn user representation from different aspects, it can capture the real preferences of users more accurately, and solve the two key problems of data sparsity and data noise at the same time to achieve better recommendation effect. Extensive comparison experiments on three publicly available datasets demonstrate the effectiveness of the MLKG model.