Attention-driven Graph-Sequence Fusion Network for Next Point-of-Interest Recommendation

Next Point-of-Interest (POI) recommendation has been widely applied on social network-based platforms, aiming to recommend a user’s potential next location based on their historical check-in data. Current research faces limitations in both graph neural networks and sequential learning models, with challenges in fully extracting user check-in features and effectively integrating global and personal information. To address these issues, we propose a model that fuses features obtained from a graph neural network and a sequential encoder through an attention fusion mechanism, followed by learning with a hierarchical Transformer encoder to enhance feature representation. Our model has been extensively tested on three datasets, achieving promising results.