Leveraging semantic techniques for social media analysis applied to disaster management

Social media are nowadays a common tool that people use to communicate. Through social media people express their feelings and exchange information. The information exchanged concerns events, places, objects etc... and reflects their interests or their psychological state. Micro-blogging has become one of the important tools in social media, where people exchange information through short messages. The volume of such messages has increased tremendously in the previous years, reaching a couple ofmillionmessages per day on somemicro-blogging services such as Twitter. The amount of posts per day can be considered a wealth of information potentially useful in case of a disaster. It is interesting to note that human beings can act as live sensors which collect and disseminate information. It is hard to extract meaningful information from the data due to several challenges, data incompleteness, huge amounts of data and absence of sound context. In this thesis, we aimed at addressing the challenges by relying on semantic techniques that can improve the context of data. We applied our techniques to crisis assessment and management (CAM) and disaster management (DM). Using social media to collect information relevant to a disaster is helpful to aid disaster management efforts. Social media provides an inexpensive means to collect data that can be used to achieve several crucial tasks in disaster management, for example, situational awareness. The set of challenges to build situational awareness can be summarized as compilation and visualization of complex information, rapid coordination of private resources, fast collection and dissemination of information during disasters. The semantic techniques developed used for situational awareness, sentiment analysis, early detection and decision support proved to yield better results than the state of the art techniques.