A Novel Multi-behavior Contrastive Learning and Knowledge-Enhanced Framework for Recommendation

Learning accurate user and item embeddings through multi behavior and knowledge graph is crucial for modern recommendation systems, which unique challenges that cannot be handled by current recommendation solutions. In particular: (1) Both multi behavior information and knowledge graph suffer from noise problem, which affects the embedding learning of users and items. (2) Leverage dependencies and complementarities between different behaviors to capture users’ more comprehensive preferences. In this work, we propose a novel Multi-Behavior Contrastive Learning and Knowledge-enhanced Framework for Recommendation (MCKR), including two subject-specific modules to address the above challenges. Specifically, we design a multi-behavior aware module to extract personalized user behavior information for user embedding enhancement, propose inter-behavior and intra-behavior contrastive learning to solve the noise problem in multi-behavior information. In addition, we design a knowledge enhancement module that uses knowledge graphs to learn item representations using both structural and semantic enhancement methods and propose a hierarchical comparative learning approach to further mitigate the noise problem associated with knowledge graphs. Extensive experiments and ablation tests on the three real-world datasets indicate our MCKR outperforms various state-of-the-art recommendation methods and verify the effectiveness of our method.