Unveiling transit mobility structure towards sustainable cities: An integrated graph embedding approach

•A transit mobility community detection problem is formulated.•Mobility communities are detected via embedding multiple-source transit data.•The proposed approach was evaluated using transit data from Shenzhen, China. Detecting urban mobility structure, i.e., segmenting urban areas into disjoint clusters with similar mobility patterns, can facilitate our understanding of how a city is organized and how different parts of a city interact with each other, which underpin informed decision-making in achieving sustainable urban transportation development and resilient society. In this study, we propose to tackle a novel transit mobility structure detection problem, which hinges on high-level mobility patterns that characterize collective movement dynamics across the study region. To this end, we propose a machine learning-based approach to discover meaningful urban mobility structure using big transit data. We contend that both transit mobility patterns and local urban function information should be considered during the detection of transit mobility structure. By integrating different sources of urban data, we model the network of transit mobility as an attributed graph: local static urban functions are described by attributed features for graph nodes whereas travel dynamics are captured via a transit mobility pattern matrix. Similarities of both attributed features and transit mobility patterns are jointly embedded to derive compact low-dimensional vector representations via graph auto-encoder. Mobility structure is then extracted using unsupervised clustering and gap statistics. The proposed approach is capable of synthesizing both mobility and static information in a data-driven manner, preserving original urban topological structure and resident movement dynamics. The proposed approach was evaluated using real-world transit data collected in Shenzhen City, China. Experimental results and analyses demonstrate that the proposed approach has the applicability of unveiling meaningful transit mobility structure in large metropolitan areas. The detected mobility community maps present a holistic overview of public transit movement structure, enabling decision makers to make informed decisions on sustainable urban development and transit management.