Beyond Trajectory Statistics: Community-Consistent Human Mobility Generation via Evolving User-Location Graphs
The authors developed MOTION, a multi-agent framework that generates synthetic human mobility by modeling dynamic user-location graphs and preserving community structures. The model uses the Leiden algorithm to extract historical interaction priors, which guide interacting user, location, and manager agents to refine trajectory decisions from both individual and community perspectives. Evaluated on two real-world datasets, MOTION outperformed existing mobility generation baselines by at least 4.05% across metrics measuring individual trajectories, spatial flow distributions, and graph structures.
Why it matters — By moving beyond individual trajectory statistics to capture evolving community-level interaction patterns, this approach produces synthetic mobility data that better reflects complex social and urban graph topologies for downstream spatial computing applications.