MoveFormer: Spatial Graph Periodic Injection Network for Next POI Recommendation

Next Point-of-Interest(POI) recommendation task is based on the user’s historical check-in sequence, aiming to recommend the next location that the user may be interested in. It plays a crucial role in location-based services. Most existing spatial knowledge modeling methods cannot simultaneously consider spatial locations’ relationship and spatial movement periodicity in check-in sequence. Additionally, these methods model user preference knowledge as a whole while neglecting certain critical components of preference and the interaction pattern among these components. To address this, we propose a Spatial Graph Periodic Injection Network (SGPN) to simultaneously model POIs’ position relationship and sequence’s spatial movement periodicity. Subsequently, we design a Preference Attention (PA) network to handle user preference from the perspectives of semantic knowledge, spatial knowledge, and frequency components, and utilize a scale expansion-aggregation structure to model the complex interaction between semantic and spatial features. Finally, we propose the MoveFormer combining the above techniques. Experimental results on two datasets demonstrate the significant superiority of our proposed method over existing state-of-the-art approaches.