Recommendation Algorithms Fusing Knowledge Graph and User Behavior

This Traditional recommendation algorithms fall short in utilizing semantic information, and popular approaches today use graph neural networks (GNNs) to mine the higher-order structure and semantic information of the knowledge graph for recommendation. However, these methods tend to ignore the representation information between items that users have interacted with in their historical interaction behavior. To address this problem, this paper proposes a recommendation model that fuses knowledge graph and user behavior (KGUB). First, the idea of contrastive learning is utilized to encode the interaction patterns of users and items to learn the embedded representations of users. Next, these user embedding vectors are used to feed back the generation of item embedding vectors, by combining the interactions of relationship vectors in the knowledge graph, thus achieving weighted aggregation of the features of the item's neighboring nodes. Finally, the obtained user and item embedding vectors are fused by cross-compression units to capture the higher-order interactions between users and items and perform click-through rate prediction. In order to verify the feasibility and effectiveness of the model, comparison experiments are conducted with five baseline models on two public datasets, and the AUC and F1 metrics outperform the other baseline models, demonstrating that the model can provide more accurate recommendation results.