| Abstract: | A growing number of citizens and local governments are using tweets to communicate during natural disasters. A good understanding of the communication contents and behaviors is critical for disaster relief. Previous work has used crisis taxonomies or manually labeling methods to understand the content. However, such methods usually require extra efforts to find insights related to specific events. In this paper, we use a semi-automatic framework to extract specific topics from the communication contents of citizens and local governments, combined with the spatiotemporal information to explore: 1) the spatiotemporal bursts of topics; 2) the change of topics with respect to the severity of disaster; and 3) communication behaviors. We use tweets collected during 18 snowstorms in the State of Maryland, US, as a case to study. The study reveals the communication differences due to the urban-rural divide or to the severity of the snowstorms. The insights suggest that local governments could potentially adapt the context of information delivered to citizens so as to match their needs. |
