Multi-Level Graph Attention Network Contrastive Learning for Knowledge-Aware Recommendation

Researchers Zhifei Hu and Feng Xia have proposed a novel multi-view graph contrastive learning framework designed to enhance knowledge-aware recommendation systems. Addressing common limitations in existing approaches, such as sparse labels, insufficient graph structure learning, and noisy entities within knowledge graphs, this new method aims to improve recommendation accuracy. The framework utilizes multi-view knowledge graph distillation to refine user representations, allowing for more precise modeling of user preferences regarding entities and relations. Additionally, it aggregates neighborhood entity information to create informative item representations. A key innovation is the multi-level self-supervised contrastive learning module, which operates across Inter-Level, Intra-Level, and Interaction-Level perspectives. This design enhances the model's ability to generalize within classes while distinguishing between them, facilitating effective multi-dimensional feature modeling. Extensive experiments conducted on three public datasets demonstrate that the proposed framework consistently outperforms current state-of-the-art methods. Ablation studies further confirm the effectiveness of each individual module within the model, marking a significant advancement in the integration of graph neural networks and knowledge graphs for information retrieval and artificial intelligence applications.