Next POI Recommendation Based on Time Slot Preferences and Bidirectional Transformation Modeling

Zhao Liu, Wei Liu, Huaijie Zhu, Jianxing Yu, Jian Yin

Abstract: Predicting the next Point-of-Interest (POI) is crucial for location-based services, offering personalized recommendations and enhancing user experiences. However, existing methods often fail to adequately capture the temporal dynamics of user movements, leading to suboptimal performance in time-sensitive recommendations. In this paper, we propose a novel model, the Time-enhanced Sequential Prediction Model (TSPM), designed to improve the accuracy of next POI recommendations by incorporating time slot preferences and bidirectional transformation modeling. We utilize a Time-enhanced Sequence-based Dynamic Graph (TSDG) that captures both the temporal transitions of POIs and the sequential dependencies in user behavior. By dividing the day into distinct time slots and embedding temporal information directly into the graph structure, TSPM effectively models the dynamic nature of user movements. Furthermore, TSPM leverages bidirectional transformation modeling to capture both incoming and outgoing transitions, aligning with user time slot preferences for more precise recommendations. Our experiments on two real-world datasets demonstrate that TSPM significantly outperforms existing methods in terms of prediction accuracy, highlighting the importance of incorporating temporal dynamics and bidirectional modeling into POI recommendation systems.