STAM: A Spatiotemporal Aggregation Method for Graph Neural Network-based Recommendation

Graph neural network-based recommendation systems are blossoming recently, and its core component is aggregation methods that determine neighbor embedding learning. Prior arts usually focus on how to aggregate information from the perspective of spatial structure information, but temporal information about neighbors is left insufficiently explored. In this work, we propose a spatiotemporal aggregation method STAM to efficiently incorporate temporal information into neighbor embedding learning. STAM generates spatiotemporal neighbor embeddings from the perspectives of spatial structure information and temporal information, facilitating the development of aggregation methods from spatial to spatiotemporal. STAM utilizes the Scaled Dot-Product Attention to capture temporal orders of one-hop neighbors and employs multi-head attention to perform joint attention over different latent subspaces. We utilize STAM for GNN-based recommendation to learn users and items embeddings. Extensive experiments demonstrate that STAM brings significant improvements on GNN-based recommendation compared with spatial-based aggregation methods, e.g., 24% for MovieLens, 8% for Amazon, and 13% for Taobao in terms of MRR@20.