Social Media Reactions Reveals Spatial and Temporal Discrepancies during Hurricanes

Abstract

Compared to other forms of data source, social media data prevails at being a better real-time agent that reflect social responses towards a hazardous event and has been widely studied. Recently, the giant leap taken by the natural language processing techniques enable in-depth analysis of the textual content of social media posts which leads to better knowledge generation. In this paper, we utilize a multi-class predicting model named VictimFinder 2.0 that can categorize tweets into fine-grained topics such as humanitarian help request, animal helps request, infrastructure damage and shelter information. The identified tweets can be later processed with an advanced geoparsing pipeline to retrieve the accurate location where the tweets are posted. Hurricane Harvey, Irma and Ian are selected as the case study and around 10 million tweets are analyzed. Harvey and Irma which took place at almost the same time are compared to reveal the discrepancies in spatial patterns while Irma and Ian, which both made landfall in Florida are compared to reveal the temporal variations. Multiple driving factors during hazardous events can be discovered. Relying on the output of this study, first responders and policy maker can make better and more informed decision when dealing with future hurricanes.