Mining the disaster hotspots - situation-adaptive crowd knowledge extraction for crisis management

When disaster strikes, emergency professionals rapidly need to gain Situation Awareness (SAW) on the unfolding crisis situation, thus need to determine what has happened and where help and resources are needed. Nowadays, platforms like Twitter are used as real-time communication hub for sharing such information, like humans' on-site observations, advice and requests, and thus can serve as a network of “human sensors” for retrieving information on crisis situations. Recently, so-called crowd-sensing systems for crisis management have started to utilize these networks for harvesting crisis-related social media content. However, up to now these mainly support their human operators in the visual analysis of retrieved messages only and do not aim at the automated extraction and fusion of semantically-grounded descriptions of the underlying real-world crisis events from these textual contents, such as providing structured descriptions of the types and locations of reported damage. This hampers further computational situation assessment, such as providing overall description of the on-going crisis situation, its associated consequences and required response actions. Consequently, this lack of semantically-grounded situational context does not allow to fully implement situation-adaptive crowd knowledge extraction, meaning the system can utilize already established (crowd) knowledge to correspondingly adapt its crowd-sensing and knowledge extraction process alongside the monitored situation, to keep pace with the underlying real-world incidents. In the light of this, in the present paper, we illustrate the realization of a situation-adaptive crowd-sensing and knowledge extraction system by introducing our crowdSA prototype, and examine its potential in a case study on a real-world Twitter crisis data set.