Spatio-Temporal Aware Knowledge Graph Embedding for Recommender Systems

Liu Yang, Xin Yin, Jun Long, Tingxuan Chen, Jie Zhao, Wenti Huang

Abstract: Knowledge Graphs (KGs) have been incorporated into recommender systems as side information to solve the clas-sical data-sparsity and cold-start problems, with explanations for recommended items. Traditional embedding-based recommender systems generally utilize abundant information from KGs directly to enrich the representation of items or users, but the influences of spatial and temporal dependencies are usually ignored among them. In this paper, we propose a Spatio-Temporal Aware Knowledge Graph Embedding (STAKGE) for recommender systems, which incorporates spatio-temporal information with bias when propagating potential preferences of users in knowl-edge graph embedding. Moreover, we construct a multi-source KG-based recommender dataset - YelpST, containing spatio-temporal information. The experiments on YelpST dataset show that our proposed approach can capture comprehensive spatio-temporal correlations and improve the prediction performance as compared to various state-of-the-art baselines.