A Context‐Enhanced Graph Neural Network Operator for Edge‐to‐Edge Learning in Human Mobility Networks

ABSTRACT Lying at the core of human–environment interactions, mobility reflects how individuals and groups are shaping and shaped by their socioenvironmental surroundings. Human mobility patterns are vital in cities, but their modeling remains challenging due to the spatial heterogeneity of locations, irregular spatial interactions, diverse travel preferences, and dynamic environmental factors. To approach these complexities, human mobility networks are often contextualized in spatially embedded networks, where places are represented as nodes and place connections are formalized as edges. Leveraging such graph structure of mobility patterns, existing node‐centric Graph Neural Network (GNN) operators have been proven effective in learning node‐context. However, mobility is not solely contextualized in nodes; edge‐context, which involves the complex spatial associations among edges, are equally important. To mitigate the gap, we introduce the Edge‐Activating Graph Operator (EAGO) as a novel GNN operator, for context‐enhanced learning in human mobility networks. EAGO retains conventional Node‐to‐Node learning while introducing an Edge‐to‐Edge Aggregation mechanism with cross‐shaped attention. Two fusion rules, that is, Edge‐to‐Node and Node‐to‐Edge, are designed in EAGO to enable bi‐directional context‐enhanced feature exchange between edges and nodes. Utilizing a large‐scale mobile positioning data set in the Twin Cities Metropolitan Area, USA, we find that EAGO outperforms existing edge‐learning methods in both flow generation and check‐in prediction tasks, and existing node‐centric operators can be easily adapted within EAGO. Further empirical findings validate the effectiveness of edge‐context and highlight the potential of edge‐to‐edge learning for future studies on human mobility networks.