Human-Informed Urban Vitality: Integrating Form, perception, and design via Heterogeneous Graph Learning

Urban vitality reflects the capacity of urban environments to attract social interaction, economic activity, and cultural exchange. Measuring and predicting vitality at fine spatial scales, such as street blocks, remains challenging due to the complex interplay of spatial form, function, and human perception. This study introduces a humaninformed framework that integrates dynamic street activity and static functional attributes into a composite Urban Vitality Index (UVI) and translates each urban block into a heterogeneous graph that captures building, street, and cross-layer spatial relationships. Using this representation, we evaluate multiple graph neural network models, including GCN, GAT, GIN, GraphSAGE, and the Heterogeneous Graph Transformer (HGT), to classify vitality levels across 3,930 blocks in Manhattan, NY. Among these, HGT demonstrates the strongest performance, highlighting its capacity to learn from complex, multi-type urban structures. The proposed framework provides a scalable pathway for linking urban form, function, and perception through heterogeneous graphs, offering valuable insights for datadriven urban planning and design.