关键词: 知识图谱, 多视角, 知识对齐, 对比学习, 门控融合

Abstract: Abstract: Knowledge graph-based recommendation methods learn representations of user and item nodes by integrating relational connections from both knowledge graphs and user-item interaction graphs to recommend suitable items. However, since knowledge graphs contain both noisy and high-quality relations, the primary challenge lies in circumventing noisy relations while effectively mining valuable relations. Existing methods typically relied on global reconstruction strategies to optimize graph relations through either pruning noisy relations or mining high-quality relations in a single manner, thereby learning user and item representations. However, global perspectives struggled to sufficiently capture detailed local information and tended to overlook potential complementarity between local and global information. Furthermore, solely employing pruning or supplementation strategies failed to simultaneously mitigate interference from noisy relations and comprehensively exploit high-quality relations. To address these issues, a multi-perspective relation-enhanced knowledge graph recommendation method was proposed. This method alleviated noise impacts in knowledge graphs and user-item interaction graphs through an item hybrid relation alignment module and an interaction hybrid relation enhancement module, while deeply exploring valuable high-order relations. The item hybrid relation alignment module extracted local and global relations separately via importance pruning strategies and high-order relation mining approaches, followed by a knowledge alignment method to synergize both types of information for effectively refining high-quality item auxiliary information. The interaction hybrid relation enhancement module constructed local hybrid pruning relation graphs and global hybrid supplementation relation graphs, with cross-channel and cross-layer contrastive learning employed to enhance their informational complementarity, thereby comprehensively learning user and item representations. Finally, hierarchical gated adaptive fusion was adopted to integrate multiple sets of user and item embeddings for recommendation. Experimental results on multiple datasets demonstrated that compared with the VRKG4REC baseline model, the proposed model achieved improvements of 1.12% to 10.16% in NDCG metrics.

Key words: knowledge graph, multi-perspective, knowledge alignment, contrastive learning, gate fusion