Résumé
"It is widely held that social media is a valuable source of information for responders to mass emergencies. However, little is known about how much useful situational information social media data streams contain and how this content varies across types of events. In this paper, we conduct a thorough investigation of the novel situational information provided by public Twitter posts produced during and immediately following three fundamentally different disasters: the 2012 Newtown Connecticut shooting, 2013 Moore tornado, and 2013 Alberta floods. Our analysis reveals that these events yield data streams that contain quite different types and amounts of situational information, generated by different kinds of individuals with varying relationships to the event of interest. Furthermore, we find that, across all events, event-specific hashtags are rarely used in tweets that carry novel information, suggesting that event-specific hashtag/keyword filters miss much novel information. Collectively, these findings highlight a number of factors that can inform how social media is used to facilitate response and recovery efforts for specific mass emergencies."--Page 1.