Discovering disaster events from social media streams

Natural and man-made disasters can both cause severe loss of lives and economic damage. Examples include earthquakes, floods, and road crashes. Nevertheless, rapidly and accurately identifying the latest status of a disaster event is undoubtedly one of the most difficult tasks for agencies in crisis management. In this work, we thus propose to monitor online data streams in social media to detect and track real-world events. Unlike conventional media, social media is advantageous because of its immediateness, huge data scale, and worldwide availability. Nevertheless, messages generated by netizens can be incomplete, subjective, or even error prone. Only with an appropriately designed scheme can invaluable clues embedded in huge amounts of online messages be discovered when carefully exploiting the information over content, temporal, and social dimensions. Specifically, we collect data from multiple social networks, conduct real-time analysis, and present interactive visualization. Experimental studies show that the proposed scheme is feasible for agencies in practice.