Multi-View Spatial-Temporal Enhanced Hypergraph Network for Next POI Recommendation

Abstract

Next point-of-interest (POI) recommendation has been a prominent and trending task to provide next suitable POI suggestions for users. Current state-of-the-art studies have achieved considerable performances by modeling user-POI interactions or transition patterns via graph- and sequential-based methods. However, most of them still could not well address two major challenges. 1) Ignoring important spatialtemporal correlations during aggregation within user-POI interactions; 2) Insufficiently uncovering complex high-order collaborative signals across users to overcome sparsity issue. To tackle these challenges, we propose a novel method Multi-View Spatial-Temporal Enhanced Hypergraph Network (MSTHN) for next POI recommendation, which jointly learns representations from local and global views. In the local view, we design a spatial-temporal enhanced graph neural network based on user-POI interactions, to aggregate and propagate spatial-temporal correlations in an asymmetric way. In the global view, we propose a stable interactive hypergraph neural network with two-step propagation scheme to capture complex high-order collaborative signals. Furthermore, a user temporal preference augmentation strategy is employed to enhance the representations from both views. Extensive experiments on three real-world datasets validate the superiority of our proposal over the state-of-the-arts. To facilitate future research, we release the codes at https://github.com/icmpnorequest/DASFAA2023_MSTHN.