Integrating Social and Knowledge Graphs in GNN-Based Recommender Systems

Zengyi Yu, Yonggu Wang, Mengfang Hu, Zihan Wang

Abstract: Graph Neural Networks (GNN) have emerged as a powerful tool in recommendation systems due to their ability to adeptly model complex relational data. Despite their potential, existing GNN-based approaches often fail to fully harness the synergistic benefits of integrating social networks and knowledge graphs into the recommendation process, over-looking the nuanced differences between these data sources. To address these gaps, we propose a novel Integrating Social and Knowledge Graphs (ISKG) framework tailored for GNN-based Recommender Systems. The ISKG model amalgamates user-item interactions, social connections, and knowledge graph insights into a unified representation, enhancing the recommendation quality through a multi-faceted approach. It starts with generating initial embeddings, progresses through a fusion layer for feature amalgamation, and refines these features in successive propagation layers. An innovative Adaptive Weighting Mechanism dynamically balances the influence of social and knowledge graph-enhanced features, leading to a Prediction Layer that finalizes the recommendations. Our comprehensive evaluation showcases ISKG’s superiority over conventional baselines, high-lighting its ability to achieve an effective balance between social and knowledge-based recommendations, thus paving the way for more accurate and nuanced recommendation systems. The project details are available at https://yuzengyi.github.io/ISKG/.