A Time Slot Tree-Based Network Model for POI Recommendation Systems

Next Point Of Interest(POI)recommendation predicts the places that a user is likely to visit in the future based on their historical check-in trajectories and current state. Most current POI recommendation models are based on sequence prediction, However, there is still a significant gap between the filtering of specific user behaviors in different time periods and the actual user decision-making behavior. Therefore, we propose a new time slot tree(TSTNet) model for user trajectory check-in and learning, aimed at extracting specific user behaviors in different time periods. TSTNet divides a day into multiple time periods to understand the user‘s preferred behaviors during different time slots. The model employs a GNN framework, integrating LSTM gating mechanisms and attention mechanisms to achieve tree-structured modeling. It better captures user preferences within specific time periods. Experimental results show that the prediction accuracy has improved on three LBSN datasets.