SR-KGELS: Social Recommendation Based on Knowledge Graph Embedding Method and Long-Short-Term Representation

Data from user-item interactions and social data can be integrated to improve the effectiveness of social recommendations. Graph neural networks (GNNs) have gained popularity in social-based recommender systems due to their inherent integration of node information and topology. However, most research has focused on how to deeply model users using various datasets, with less emphasis on item relationships. Furthermore, users’ changing interests over time, preferences in long-term patterns, and differences in rating behavior across perspectives have received little attention. In this work, we propose a social recommendation based on knowledge graph embedding method and long-short-term representation (SR-KGELS). The SR-KGELS learns user and item features by combining long- and short-term representations and using attention mechanisms to discern the strength of heterogeneity associated with social and relevant relationships. In addition, we treat the differences in users’ scoring behaviors as a relative location difference problem in the embedding space, and model it with a knowledge graph embedding method called TransH to improve the generalization ability of the main rating model. Experiments on two real-world recommender system datasets validate the effectiveness of SR-KGELS.