Harnessing Rich Multi-Modal Data for Spatial-Temporal Homophily-Embedded Graph Learning Across Domains and Localities

Modern cities are increasingly reliant on data-driven insights to support decision making in areas such as transportation, public safety and environmental impact. However, city-level data often exists in heterogeneous formats, collected independently by local agencies with diverse objectives and standards. Despite their numerous, wide-ranging, and uniformly consumable nature, national-level datasets exhibit significant heterogeneity and multi-modality. This research proposes a heterogeneous data pipeline that performs cross-domain data fusion over time-varying, spatial-varying and spatial-varying time-series datasets. We aim to address complex urban problems across multiple domains and localities by harnessing the rich information over 50 data sources. Specifically, our data-learning module integrates homophily from spatial-varying dataset into graph-learning, embedding information of various localities into models. We demonstrate the generalizability and flexibility of the framework through five real-world observations using a variety of publicly accessible datasets (e.g., ride-share, traffic crash, and crime reports) collected from multiple cities. The results show that our proposed framework demonstrates strong predictive performance while requiring minimal reconfiguration when transferred to new localities or domains. This research advances the goal of building data-informed urban systems in a scalable way, addressing one of the most pressing challenges in smart city analytics.

💡 Research Summary

This paper presents a novel framework designed to tackle one of the most pressing challenges in smart city analytics: the integration and effective utilization of heterogeneous, multi-modal urban data for scalable and generalizable predictive modeling across different domains and geographic localities.

The core problem addressed is the fragmentation of city-level data, which is often collected independently by various agencies in diverse formats and standards. While national datasets offer broad coverage, they are highly heterogeneous. The proposed solution is a hierarchical data pipeline that fuses two categories of data: “Features” (globally/nationally available data like demographics, economic indicators, land cover, and weather) and “Observations” (locally-sourced, domain-specific time-series data such as ride-share requests, crime reports, and traffic crashes). These are processed into 1D, 2D, and 3D dataset classes to feed into a graph learning model.

The key technical innovation is the introduction of a “homophily-embedded graph” structure for spatial-temporal Graph Convolutional Network (GCN) learning. Traditional GCNs for urban problems often rely on adjacency matrices based solely on Euclidean distance between geographical regions (e.g., census tracts). The authors argue this is insufficient as it ignores socio-economic and environmental similarities between areas. Their method enriches this distance-based adjacency matrix (A_d) by weighting it with a composite correlation matrix. This correlation matrix is derived from 48 carefully selected features spanning demographics, land use, and POI/economic activity, capturing the multifaceted “homophily” or similarity between regions. The resulting adjacency matrix (A’) more accurately reflects real-world urban dynamics where proximate areas can be dissimilar and distant areas can share common characteristics.

The framework’s effectiveness and generalizability are rigorously evaluated through five real-world case studies across three domains (ride-sourcing demand, crime incidence, traffic crashes) and three distinct U.S. cities (Chicago, Pittsburgh, Oakland). The experiments demonstrate that models utilizing the proposed homophily-embedded graph consistently outperform those using standard distance-based graphs. Furthermore, the framework shows strong transfer learning capabilities; models pre-trained on data from one city require only minimal reconfiguration (primarily swapping the local “Observation” data) to achieve robust predictive performance in a new city or on a new domain.

In conclusion, this research advances the field by moving beyond simple spatial proximity in urban graph learning. It provides a scalable, flexible pipeline for fusing rich multi-modal data and a principled method (homophily-embedding) to inject crucial contextual socio-economic and environmental information into the graph structure itself. This work marks a significant step toward building truly data-informed urban systems that can adapt to diverse challenges and locations.