Classification and severity assessment of disaster losses based on multi-modal information in social media

Abstract

Capture and fine-grained classification of disaster loss information, combined with severity assessment are essential for emergency management departments to carry out effective emergency rescue measures. With the rapid development of social media platforms, the vast amount of user posts on social media provides critical disaster loss information during disasters. In this study, a fine-grained multimodal information-based disaster losses classification (MIDLC) model is proposed to identify the different types of disaster losses from massive social media data. This model uses three datasets, i.e., microblogging posts, government announcements about public events, and microblogging images. The disaster losses are divided into five types, i.e., casualties, houses and buildings collapse, municipal infrastructure damage, public service facilities damage, and impact on production and daily activities. Subsequently, this study proposes a disaster loss severity assessment system to evaluate the severity of different types of disaster loss reflected by social media, guiding targeted rescue response activities. The severity assessment system measures the severity from four dimensions: event information characteristics, information dissemination strength, official response, and user emotional volatility. Finally, the proposed MIDLC model and the severity assessment system are illustrated by investigating three disaster events. Results show that the MIDLC model proposed in this study significantly improves the performance of disaster losses classification. In these five disaster loss types, positive correlation exists between the casualty losses and houses and buildings collapse losses.