Detecting temporal socioeconomic changes of regions in Chicago via urban region representation learning

Abstract

Urban region representation learning aims to extract valuable insights for understanding urban dynamics from diverse and intricate urban data. Since human mobility is greatly related to the socioeconomic status in a city, existing studies have comprehended urban regions through human mobility. However, it is difficult to analyze temporal variation through the previous region representations, and intrinsic evaluation is also required. Therefore, this study aims to detect temporal socioeconomic changes in urban regions via region representation learning. We modify the HUGAT model to represent region embedding for all years in one latent space. Our model outperforms baselines in predicting socioeconomic levels compared to state-of-the-art models. Furthermore, we propose a temporal similarity measure to analyze the temporal variation between 2013 and 2020 and discover regions where socioeconomic changes happen due to the inflow or outflow of a social class. This result enables efficient urban planning by monitoring each region that undergoes different changes due to social phenomena such as racial segregation or gentrification in Chicago.