Integrating Social and Knowledge Graphs with Time Decay Mechanisms

Can Lin, Xiaoyang Han, Zengyi Yu, Jiaxin Du

Abstract: This paper introduces an innovative approach to recommender systems by integrating social connectivity enhancement, knowledge graph augmentation, and a temporal decay mechanism within a Graph Neural Network (GNN) architecture. Our proposed model, ISKG-TD, aims to capture the dynamic nature of user preferences by incorporating temporal information into the user-item interactions, thereby addressing the limitations of static models. By leveraging the social graph for user connections and the knowledge graph for item attributes, our model enhances the recommendation quality with rich contextual information. The integration of a time decay mechanism allows for the differentiation of interactions over time, ensuring the model remains sensitive to the evolving nature of user preferences. We conduct comprehensive experiments on sampled Yelp dataset, demonstrating the superiority of ISKG-TD over traditional and state-of-the-art models in terms of recommendation accuracy and relevance. The results highlight the effectiveness of our multi-graph approach and the potential of time-sensitive modeling in recommender systems. The code is available at https://github.com/codelover123hxy/ISKG_TD.