Modeling the place resilience with heterogeneous graph neural networks

Keywords: Artificial Intelligence, Disaster, Graph neural network, Resilence, POI
Abstract Type: Paper Abstract

Abstract

Resilience is the ability to bounce back to normal after disaster hits. This ability is related to complex reasons and draws much research attention. In this research, we measure the place's resilience using heterogeneous data such as visitation patterns, demographic information, and place of interest (POI). We proposed a heterogeneous graph neural network to model the resilience of places. Our experiments in Galveston, TX, USA, showed that our model could better predict the resilience of places compared to regression models and other homogeneous baselines. Further analysis revealed distinct patterns of resilience related to visitation numbers, visitors' travel behavior, and geographical information. Compared with resilience research based solely on the visit count, our method catches more information and derives comprehensive knowledge about the place's resilience. Our proposed heterogeneous graph neural network would be an objective and rigorous tool to analyze place resilience.