Enhanced Knowledge Graph Recommendation Algorithm Integrating Graph Attention and Contrastive Learning

Aiming at the problem that current knowledge graph based recommendation systems cannot capture complex high-order entity relationship information, an enhanced knowledge graph recommendation algorithm model (GAKR-GCL) that integrates graph attention and contrastive learning is proposed. GAKR-GCL improves the effectiveness of knowledge graph embedding modules by comparing learning and graph attention networks, while utilizing adaptive gating mechanisms combined with graph attention networks to generate more accurate item feature vectors. Specifically, firstly, the model utilizes a multi-layer perceptron to extract high-order latent features of users. Then, the model dynamically adjusts the weights of collaborative filtering vectors and knowledge graph entity vectors through an adaptive gating mechanism, generating more accurate item representations. Next, the knowledge graph embedding module is combined with graph attention network to aggregate multi hop neighborhood information, generate enhanced entity representations, and introduce multi view contrastive learning to improve the model’s generalization ability. In addition, L2 regularization is used to prevent over-fitting of the model. Finally, the normalized inner product operation is used as the evaluation function to predict the user’s preferences for the item. In three publicly available datasets, MovieLens-1M, Book Crossing, and Last.FM show that the GAKR-GCL model can effectively improve recommendation performance compared to other benchmark models.