Emergency Event Detection in Twitter Streams Based on Natural Language Processing

Real-time social media usage is widely adapted today because it encourages quick spreading of news within social networks. New opportunities arise to use social media feeds to detect emergencies and extract crucial information about that event to support rescue operations. A major challenge for the extraction of emergency event information from applications like Twitter is the big mass of data, inaccurate or lacking metadata and the noisy nature of the post text itself. We propose to filter the real-time media stream by analysing posts seriousity, extract facts through natural language processing and group posts using a novel event identification scheme. Based on a manually tagged social media feed corpus we show that false or missed alarms are limited to posts with highly ambiguous information with less value for the rescue units.