Understanding social media data for disaster management

Abstract

Social media data are increasingly being used in disaster management for information dissemination, establishment of situational awareness of the “big picture” of the disaster impact and emerged incidences over time, and public peer-to-peer backchannel communications. Before we can fully trust the situational awareness established from social media data, we need to ask whether there are biases in data generation: Can we simply associate more tweets with more severe disaster impacts and therefore higher needs for relief and assistance in that area? If we rely on social media for real-time information dissemination, who can we reach and who has been left out? Due to the uneven access to social media and heterogeneous motivations in social media usage, situational awareness based on social media data may not reveal the true picture. In this study, we examine the spatial heterogeneity in the generation of tweets after a major disaster. We developed a novel model to explain the number of tweets by mass, material, access, and motivation (MMAM). Empirical analysis of tweets about Hurricane Sandy in New York City largely confirmed the MMAM model. We also found that community socioeconomic factors are more important than population size and damage levels in predicting disaster-related tweets.