Point-of-interest (POI) recommendations provide personalized location suggestions based on user historical behavior and preferences. While current state-of-the-art methods use graph-based and sequential learning approaches to model user-POI interactions, they face two main limitations: neglecting personalized spatial-temporal features that reflect users’ periodic interests and failing to fully exploit the higher-order collaborative signals between user sequences. This study introduces a spatial-temporal attention hypergraph (STA-Hyper) model for the next point-of-interest recommendation task, which solves the challenge of capturing complex spatial-temporal correlations and higher-order collaborative signals in user check-in data. The STA-Hyper model effectively simulates the relationship between POIs and user behaviors across regions and time periods by constructing a hypergraph based on user trajectories. The model employs a multiattention mechanism to enhance the information transfer process within the hypergraph for a more accurate prediction of user preferences. Extensive experiments on three large real datasets show that the STA-Hyper model outperforms the benchmark approach and achieves state-of-the-art performance. Ablation studies further validate the effectiveness of spatial and temporal modules in improving the precision of recommendations.