Next POI Recommendation Using Higher-Order Travel Patterns

Abstract

With the rapid development of location-based social networks (LBSNs) and the growing scale of user check-in data, the next POI recommendation has become a key technology for enhancing user experience and increasing service provider value. However, traditional methods have shown limited performance in capturing collaborative relationships between user check-in sequences. The graph-based model, while capable of modeling such relationships, struggles to effectively extract higher-order information from check-in sequences. Although recent advances in hypergraph-based research offer new directions for obtaining higher-order information, existing hypergraph-based methods often perform poorly on datasets with smaller data volumes. To address these challenges, this study proposes a Higher-Order and Global Information Fusion Model. This model leverages two-hop subsequences in user check-in data to uncover higher-order relationships between sequences while preserving the ability to model global user check-in patterns. Specifically, the proposed model extracts two-hop subsequences from user check-in sequences to construct a higher-order graph, capturing frequent and periodic patterns in user behaviors. Simultaneously, inspired by the concept of global trajectory flow graphs, the model captures POI transition patterns from a global perspective by modeling users’ movements between POIs. Building on this foundation, the model integrates key spatio-temporal information, including user preferences, POI categories, and check-in times. By effectively combining local and global representations of POIs and incorporating rich spatio-temporal context, the model demonstrates high accuracy on real-world datasets, especially in cold-start scenarios. Our experiments on real-world datasets validate the effectiveness of combining higher-order relationships derived from two-hop subsequences with global POI transition patterns captured by the trajectory flow graph, highlighting their importance in next POI recommendation tasks. This study provides new insights and valuable references to more comprehensively characterize user mobility behaviors and develop advanced POI recommendation models.