Mining Twitter Data for Disaster Resilience at Two Spatial Scales

Keywords: disaster resilience, Hurricane Sandy, social media, Twitter use, sentiment analysis
Session Type: Virtual Paper Abstract
Day: Saturday
Session Start / End Time: 2/26/2022 08:00 AM (Eastern Time (US & Canada)) - 2/26/2022 09:20 AM (Eastern Time (US & Canada))
Room: Virtual 3

Abstract

Disaster resilience describes the ability of a community to bounce back from disaster impacts by resilience building activities. Social media provides an innovative way to observe human attitudes and responses, especially during disasters. However, most previous social media and disasters studies were conducted at a coarse spatial scale such as by county. This study analyzes Twitter activities during Hurricane Sandy from October 23 to November 12, 2012 at both the county and the zip-code levels in the five affected states. The study examines two questions: (1) will the relationships between disparities in social media use and disparities in disaster resilience found at the county level in previous studies still hold at the zip-code level? And (2) what new information or patterns can be revealed with the zip-code level analysis? Results show that the correlations between Twitter use indices and social-environmental variables representing community resilience found at the county level in previous studies still hold, but the correlations are weaker when the study is conducted at the zip-code level. On the other hand, new and useful patterns are revealed using data at the zip-code level. Results show that zip codes that have major transportation hubs, commercial and tourist activities, or low night-time population are major factors affecting Twitter use indices and hence the correlations. Future research should consider adding land use types and population dynamics to help improve the use of social media for disaster resilience analysis. Furthermore, a multiscale analysis approach can reduce the uncertainties in spatial and relationship analysis.