Proceedings of the Annual Meeting of the Cognitive Science Society

In recent years, with the widespread popularity of location-based social network platforms, the data generated by users on social networks has grown exponentially. There has been a growing focus on the problem of POI (Point-of-Interest) recommendations. Unlike traditional sequence recommendation that primarily considers the temporal dimension, POI recommendation needs to account for the influence of geographical information to a large extent. However, previous works in the graph construction process often only consider the places users have visited, neglecting those they haven't been to. To address this, we propose a Dual Weighted Graph Convolutional Network for POI recommendation called DualPOI. Specifically, we first leverage graph neural networks and attention mechanisms to capture users' local trajectory preferences for visited POIs. A delicately designed spatiotemporal encoder is conducted to model users' local spatiotemporal preferences. Subsequently, using a dual graph convolutional approach, we transfer the user's local preference information to a global scope, thereby modeling novel preferences for unvisited locations. Extensive experiments on four real-world datasets validate the effectiveness of our proposed method in enhancing the accuracy of POI recommendations. Comprehensive ablation studies and parameter analysis further confirm the efficacy of the proposed modules.