Memory-Enhanced Period-Aware Graph Neural Network for General POI Recommendation

Tianchi Yang, Haihan Gao, Cheng Yang, Chuan Shi, Qianlong Xie, Xingxing Wang, Dong Wang

Abstract: As the popularity of Location-based Services increases, Point-of-Interest (POI) recommendations receive higher requirements to characterize the users, POIs and interactions. Although many recent graph neural network-based (GNN-based) studies have tried working on temporal and spatial factors, they still cannot seamlessly handle the temporal locality and spatial consistency. To tackle this issue, we propose a novel Memory-enhanced Period-aware Graph neural network for general POI Recommendation (MPGRec). Specifically, it exploits the advantages of the GNN module in characterizing user preferences. Moreover, we develop a period-aware gate mechanism after the GNN information propagation to characterize the temporal locality, and devise a dynamic memory module to extract, store and disseminate global information for spatial consistency. Furthermore, we propose a reading and writing strategy to merge the GNN module and memory module into a unified framework. Extensive experiments are conducted on four real-world datasets, and the experimental results demonstrate the effectiveness of our method.