Aggregating users’ preferences for points of interest (POIs) to infer regional-level mobility patterns presents a crucial challenge in region of interest (ROI) recommendation. However, existing methods mostly overlook dependencies of user behavior within a region and fail to adequately consider the adaptability of heterogeneous relationships between users and ROIs. Therefore, we propose a collective user mobility pattern guided ROI recommendation method (CM-ROI). A spatiotemporal behavior graph incorporating dynamic subgraphs is constructed. It mines common features of user behavior preferences and extracts collective user mobility patterns, thereby addressing the insufficiency in mining behavioral dependencies from a collective perspective. Subsequently, local interaction features and global association features are extracted via a co-attention mechanism and an adaptive meta-path generation mechanism. The latter is achieved by mining the multivariate synergy among different node types, effectively enhancing the adaptability of heterogeneous relationship modeling. Finally, user preferences are predicted through a multilayer perceptron by integrating the two types of relational features, enabling personalized ROI recommendation. Experiments on two real-world datasets show that CM-ROI consistently outperforms state-of-the-art baselines across multiple metrics.
Wang et al. (Mon,) studied this question.