A Heterogeneous Attention Network Model for Longitudinal Analysis of Socioeconomic and Racial Inequalities in Urban Regions: Evidence from Chicago, IL

Chicago, IL has been undergoing significant demographic and racial shifts as a result of gentrification and displacement. To resolve the risk, it is necessary to monitor the socioeconomic characteristics of regions and detect regions with drastic socioeconomic changes, such as gentrification. Region representation learning using mobility data is a rapidly developing field for inferring regional demographic features. However, past approaches had a limitation in identifying longitudinal changes in urban regions. In this paper, we propose a novel approach to model the longitudinal evolution of urban regions using heterogeneous information networks and heterogeneous graph attention networks. From the experimental results on 77 community areas in Chicago, the proposed model outperforms the state-of-the-art models in predicting the racial population percentage, unemployment rate, and poverty rate. Further, we analyze the change in the similarity between region representations and detect a region where displacement of the black population is extreme due to gentrification. Our proposed approach can contribute to achieving more equitable and sustainable cities by providing a useful basis for modeling and monitoring the evolution of urban regions.