A Comparative Study of Static and Dynamic Models for Spatiotemporal Community Detection in Urban Mobility Networks

Abstract

Revealing the dynamic evolution of traffic community structure is essential for understanding shifts in residents’ spatiotemporal travel patterns. However, most studies rely on static frameworks, which cannot detect changing traffic communities in dynamic mobility environments. In this context, a dynamic spatial network growth model is applied to detect traffic communities and a comparison with the static fast unfolding algorithm is conducted using large-scale taxi trajectory data. Based on the spatial autocorrelation results of Moran’s I, we employ the fast unfolding algorithm to detect communities in large networks and the spatial network growth model to reveal how spatial interaction influences community formation. Using a large GPS dataset of Xi’an taxi trajectories collected from February 28 to March 30, 2019, the fast unfolding algorithm identifies stable, high-density communities concentrated in commercial districts and major transport hubs. In contrast, the spatial network growth model captures weekly shifts in community boundaries. By comparing five metrics of the community detection performance, including modularity, weighted embedding, centrality index, cost index, and weighted conductance. Results demonstrate that the fast unfolding algorithm is effective for identifying persistent high-flow patterns, whereas the spatial network growth model better reflects time-varying demand. Robustness and sensitivity analyses further show that the community partitions remain highly consistent, with normalized mutual information (NMI) values consistently exceeding 0.85 across different parameter settings. Overall, it can be concluded that the two models capture complementary characteristics of community structure, providing deeper insights for the analysis of spatiotemporal travel patterns and further transportation planning.