Near-real-time earthquake-induced fatality estimation using crowdsourced data and few-shot large-language models

Abstract

When a damaging earthquake occurs, immediate information about casualties (e.g., fatalities and injuries) is critical for time-sensitive decision-making by emergency response and aid agencies in the first hours and days. Systems such as the Prompt Assessment of Global Earthquakes for Response (PAGER) by the U.S. Geological Survey (USGS) were developed to provide a forecast of such impacts within about 30 min of any significant earthquake globally. However, existing disaster-induced human loss estimation systems often rely on early casualty reports manually retrieved from global traditional media, which are labor-intensive, time-consuming, and have significant time latencies. Recent approaches use keyword matching and topic modeling to identify human casualty-relevant information from social media but tend to be error-prone when dealing with complex semantics in multi-lingual text data and parsing dynamically changing and conflicting human death and injury numbers shared by various unvetted sources in social media platforms. In this work, we introduce an end-to-end framework to significantly improve the timeliness and accuracy of global earthquake-induced human loss forecasting using multi-lingual, crowdsourced social media. Our framework integrates (i) a hierarchical casualty extraction model built upon large language models, prompt design, and few-shot learning to retrieve quantitative human loss claims from social media, (ii) a physical constraint-aware, dynamic-truth discovery model that discovers the truthful human loss from massive noisy and potentially conflicting human loss claims, and (iii) a Bayesian updating loss projection model that dynamically updates the final loss estimation using discovered truths. We test the framework in real-time on a series of global earthquake events in 2021 and 2022 and show that our framework effectively automates the retrieval of casualty information faster but with comparable accuracy to those now retrieved manually by the USGS. The code associated with this work is made available at: https://github.com/SusuXu-s-Lab/Hierarchical-Earthquake-Casualty-Information-Retrieval