Disaster-news datasets for multi-label document classification, sentence classification, and abstractive document summarization tasks

Abstract

Mining of disaster-news articles can deliver highly useful information for the authorities in critical decision making and also for awareness dissemination among people, at disaster situations. A set of disaster-news articles containing more than seven thousand and six hundred news documents on COVID-19, storm, flood, heavy rain, cloudburst, landslide, earthquake, and tsunami is put forward for interested researchers to explore. It includes more than three thousand news articles only on COVID-19 considering it a disaster event. It also includes more than 4.5 thousand articles on natural disaster events prevalent in India. A dataset has been prepared for the sentence classification task from the COVID-19 articles. An abstractive summarization dataset has also been prepared for the task of automatic generation of a suitable headline for a disaster-news article. It is done by considering the articles and their titles as text and summaries. Another dataset has been prepared by combining COVID-19 and the natural disaster articles for three tasks; first, identification of the events in an article, second and third, identification of the sentences containing disaster-location and disaster-impact information respectively. The Precision, Recall, F-measure, and Accuracy scores after applying the Random Forest classifier on both datasets are presented in this paper that show impressive results.