Disaster Network Evolution Using Dynamic Clustering of Twitter Data

Ad hoc smartphone networks can be used to augment communications degraded by disasters provided that the individual ad hoc clusters can reach some "connection gateways" to get out to the Internet. This capability can be provided by devices in the surrounding area that retain cellular connectivity in addition to the connectivity provided by the specially deployed emergency equipment, if any. The disconnected areas may not be known until they are back online; however, we need a mechanism to estimate them so that the gateway devices can be best recruited to provide the connectivity. This needs to be done in a dynamic environment because of the significant mobility in the wake of the disaster. In this paper, we propose a mechanism to estimate regions that are likely to be dense but disconnected, and with significant connected devices in and around them. Such regions are most likely to benefit from the ad hoc network. Because of the lack of direct information on people (or smartphone) density, we attempt to do this by analyzing the twitter data. We use our approach on the twitter data available on hurricane Sandy in 2012.