This study presents a graph-based framework for managing urban transport infrastructure and identifying potential mobility hubs within the Seoul metropolitan rail network in the Republic of Korea. The approach integrates traditional network centrality analysis with graph neural networks to capture both structural influence and flow mediation. Using operational schedule data, Bonacich power and random-walk betweenness centrality were embedded into a graph learning model to evaluate node importance beyond the limitations of conventional shortest-path assumptions. The results reveal that top nodes exhibit high multimodal potential, acting as strategic connectors within the metropolitan transit structure. By linking network-derived hub scores with public bicycle usage data, the analysis identifies spatial overlaps between structural centrality and micromobility demand. These findings support road space reallocation and multimodal integration strategies that enhance sustainable and inclusive accessibility. The proposed framework expands the methodological scope of transport planning by combining network science and machine learning. It provides a data-driven basis for infrastructure management and policy development toward low carbon dioxide, human-centred and resilient mobility systems. The proposed framework expands transport planning by combining network science and machine learning, providing a data-driven basis for infrastructure management, mobility hub planning and policy development toward low carbon dioxide and resilient urban transport systems.
Jeong et al. (Mon,) studied this question.