Social Media Data Analysis, a Tool for Public Health Emergency Management During Natural Disasters

Annually, thousands of people suffer as a result of natural disasters. They are left without power, water, or trapped by landslides or floods, while the world follows the communities’ ravages on social networks. All of the shared information and updates related to natural disasters on social media platforms present an opportunity for data collection to help in the emergency response cycle’s preparedness and response phases. Three studies were conducted to explore the application of social media data analysis for public health practitioners during emergency responses. First, a systematic review was completed to study social media use during emergency response to natural disasters. Results indicated that social media served as a broadcasting tool for emergency warnings. It also helped determine the location of messages and construct maps to help with relief efforts. Second, an imputation method for social media users’ location was designed using social network connections to schools and school districts in Georgia. The method identified users with unrealistic or non-local locations in their profile and assigned a location using the school or school district account they follow. Tweets posted during Hurricane Matthew were used as a case study to validate the imputation method and understand the behavior of social media users in Georgia during a natural disaster. Third, content analysis with latent Dirichlet allocation models and sentiment analysis categorized hurricane-related tweets to identify the needs and sentiment changes over time. A hurdle regression model was applied to study the association of retweet frequency and content analysis topics. Content analysis conveyed that users residing in counties affected by Hurricane Matthew posted tweets related to preparedness, awareness, and evacuations, with predominantly negative sentiment. Tweets posted by those in the hurricane path during the preparedness and response phase had a lower probability of being retweeted than those outside the track (Adjusted-Rate Ratio, aRR=0.09 (95%CI: 0.09, 0.09), aRR=0.26 (95%CI: 0.26, 0.26), respectively). Results from this research can help guide data collection and analysis during emergency response. It presents an opportunity for public health emergency responders to identify damage and individuals in need of assistance with a straightforward and easy to follow approach