Laplacian-Constrained Community Detection on Mobility Graphs: A Case Study of Xiamen City, China

We propose a novel method for detecting communities in mobility graphs by leveraging the Laplacian matrix’s inherent characteristics. Our method, unlike other partitioning techniques, constructs a block diagonal similarity graph that directly uncovers the latent community structure with a predetermined number of components, which eliminates the need for potentially unreliable post-processing procedures. Applying this methodology to traditional taxis and online ride-hailing services in Xiamen, China, we reveal unique community structures within their origin-destination (OD) graphs. Notably, ride-hailing services exhibit significant spatial segregation with well-defined community boundaries. In contrast, taxis serve more cross-community trips, indicating a divergent usage pattern. To further explore trip-purpose variations across communities, we employ Poisson regression analysis. Our findings indicate that specific trip purposes such as residence, healthcare, education, and government-related activities show spatial differentiation delineated by community boundaries. However, dining and transfer demands contribute universally to trip occurrences, surpassing community boundaries.