Abstract:
There is rapid increase in the number of microbloggers everyday. Users can share their thoughts instantaneously. In this thesis, we have studied on identi cation of emergency related contributions in social media and mobile emergency noti cation. Our goal is to notify interested people immediately or beforehand about possible existence of an emergency situation. We are extracting emergency related information from microblogging systems with the help of supervised learning methods. Afterwards, we estimate the location of the incident from the messages we have collected from microblogging systems. Once we identify an incident occurrence, we send a noti cation to people nearby the incident via mobile application. We try to answer if contributors share emergency related information on microblogging systems and if so, how contributors express di erent types of incidents (ex: re, earthquake) in microblogging systems. Whatever the incident type, location information is critical to determine. Our study exposes that, in case of emergency situations contributors share location information in post. In case contributor speci es an incorrect location in pro le, we have found that in post location indicator points to the location incident occurred. We propose a model that fetches incident related posts from microblogging system, estimates incident location and noti es people nearby incident occurrence. We describe formal structure of our model and prototype developed. We evaluate our system by the accuracy of location, time estimation for incident detection about re and earthquake incidents. Once we compare detections of our system with trustworthy sources, out of 246 earthquake incidents 25, out of 85 re incidents 42 of them generated noti cations. As a result of evaluation, we assess future improvements of our system.
