A Multi-stage Classification Framework for Disaster-Specific Tweets

Abstract

The paper aims to develop a pipeline for the classification of tweets generated at the time of disasters for relief work. It is essential to first determine the types of incidents that such real-time tweets refer to. We further need to extract useful and actionable information from the streaming posts. Classifying them on the basis of contained information can facilitate work done by concerned authorities in taking immediate action. We evaluate a range of classification and feature extraction methodologies to accomplish these tasks. We also modify a public dataset of multiple events to propose a set of categories in which the information conveyed by tweets may fall into. We finally analyze and propose the best-found methodologies to extract critically informative and actionable tweets during the disasters. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd 2021.