Two-Stage Trajectory Generation Model for Realistic Human Mobility Simulation.

Human mobility data is crucial for understanding mobility patterns and provides valuable insights for fields like urban planning and tourism recommendation. To address limitations in trajectory generation methods that treat all points equally, this study proposes a two-stage trajectory simulation model that generates separate sequences for stopping points and moving points. By differentiating between these two types of points, the model can effectively capture mobility intentions and moving paths. The stopping point sequence generator utilizes a GAN-based approach to capture mobility intentions and patterns, while the moving point generator constructs a weighted directed road network graph to determine efficient paths between stopping points. Extensive experiments using a real-life mobility dataset demonstrate the effectiveness of the proposed model in generating realistic trajectories. Moreover, when used for next location prediction task, the synthetic trajectory data generated by our model produces results that closely resemble real-world situations.