Online social network sensors for influenza outbreaks

Abstract

Previous research has shown strong correlations between postings on the online social network Twitter where users complain of influenza-like symptoms, and clinical data on actual influenza rates. In addition, previous research has shown that more popular individuals in a real-life social network are infected with influenza earlier than average individuals. We collect all flu-related tweets during the 2012-2013 influenza season in order to compare the timing of flu-related tweets from more popular users compared to less popular users. No difference is seen in flu tweet timing between Twitter users with a high number of followers compared to users with a low number of followers. Restricting the Twitter network to bidirectional edges (mutual followings) performs slightly better, but is still not significant. Future work should focus on identifying edges in online social networks that indicate that two users regularly come into close physical proximity.