Intelligent and real-time alert model for disaster management based on information retrieval from multiple sources

Abstract

During the various catastrophic events of recent years, the use of social media to communicate timely information in crisis periods has become a common practice, allowing affected population to quickly publish a considerable amount of disaster information which can help managers making correct and quick decisions. In this paper, we propose a new real-time alert model for the management of natural or anthropogenic disasters. This model is based on a semi-supervised inductive technique to use unlabeled multi-source data, which are often abundant during a crisis event, with less data previously labeled than previous event. We use two sets of real-world crisis data from Facebook and Twitter manually tagged to launch streaming retrieval of relevant content: it is used for evaluating our proposed approach. Preliminary results are satisfactory.