Real-time disaster event extraction from unstructured text sources

Emergency response has to start within seconds of the onset of a disaster. However, getting a quick understanding of the situation as it unfolds is a challenging task for responding organisations due to time-consuming and labour-intensive traditional approaches. The rapid advancement of technology has provided non-traditional ways of getting to know more information, such as social media and online news. Automatically extracting events from the unstructured text can address the challenge of information scarcity currently faced by emergency responders. The task is to identify the main event from online text sources such as online news and tweets by answering 5W1H questions (who did, what, when, where, why and how). In this poster, we present our early work of detecting disaster events in real-time from unstructured text sources to improve the situation awareness of emergency responders. We have collected news headlines from the National Institute of Water and Atmospheric Research (NIWA) about historic weather events in New Zealand over a period of more than 20 years from 01.01.2000-01.06.2020. Also, we incorporated human-labeled tweets collected from the 2015 Nepal earthquake and the 2013 Queensland floods downloaded from crisisNLP . We propose a novel algorithm that uses syntactic and domain-specific rules to automatically extract the relevant phrases from news articles and tweets to provide answers to the 5W1H questions. The algorithm identifies disaster events from unstructured text sources in real-time by providing answers to these questions