A Hypergraph Structure-Based Aggregation Network for Next POI Recommendation
Zhen Zhang, Jun Chen, Xiangguo Zhao, Yaqi Gong, Jianyu Ren, Jintao Ouyang · IEEE Access · 2024
The next Point-of-Interest (POI) recommendation can effectively help users find places they are interested in, which is one of the important applications of location-based social networks (LBSNs). Recently, graph-based methods have become a popular research topic for the next POI recommendation. However, existing graph-based methods have ignored user interaction information and prior knowledge when learning context features, which limits the expression ability of the features. In addition, inadequate learning of user preferences and the oversight of noise in information aggregation contribute to inaccuracies in user representation. In this paper, we propose a hypergraph structure-based aggregation network model (HS-AGN) for the next POI recommendation. Firstly, we construct multiple user interaction hypergraphs and express prior knowledge through inner-hyperedges. We then use the mutual attention mechanism between nodes and internal hyperedges in the hypergraph to learn contextual feature embeddings with user interaction information, structural information, and higher-order relations. Furthermore, we designed a sampling probability function to aggregate user preferences and maximize the utilization of various contextual information by sampling aggregation while avoiding the impact of noisy data. Finally, we conducted extensive experiments using two real-world datasets. The experimental results demonstrate that our model consistently outperforms the state-of-the-art models.