Edge Activating Module: Learning edge-To-edge features for mobility flow generation

Human mobility is critical in urban planning, transportation systems, and public health policies. Modeling human mobility patterns is complex due to the variability in movement behaviors and the intricate spatial-Temporal dependencies involved. Graph Neural Networks (GNNs) have emerged as a powerful tool for processing non-Euclidean data structures, especially complex human mobility networks. However, existing GNN operators primarily focus on node features while underutilizing edge features, the latter contain valuable spatial relationships between locations. In this work, we propose an Edge Activating Module (EAM) that emphasizes the role of Edge-To-Edge features in human mobility networks. EAM introduces an Edge-To-Edge Attention mechanism to capture spatial structures of edges and integrates novel Edge-To-Node and Node-To-Edge fusion strategies to effectively integrate edge and node features. Our experimental results on real-world human mobility datasets demonstrate the effectiveness of EAM in improving the accuracy of flow generation tasks.