GRACE: Generating Cause and Effect of Disaster Sub-Events from Social Media Text

Abstract

In recent years, social media has emerged as a pivotal source of emergency response for natural disasters. Causal analysis of disaster sub-events is one of crucial concerns. However, the design and implementation of its application scenario present significant challenges, due to the intricate nature of events and information overload. In this work, we introduce GRACE, a system designed for generating the cause and effect of disaster sub-events from social media text. GRACE aims to provide a rapid, comprehensive, and real-time analysis of disaster intelligence. Different from conventional information digestion systems, GRACE employs event evolution reasoning by constructing a causal knowledge graph for disaster sub-events (referred to as DSECG) and fine-tuning GPT-2 on DSECG. This system offers users a comprehensive understanding of disaster events and supports human organizations in enhancing response efforts during disaster situations. Moreover, an online demo is accessible, allowing user interaction with GRACE and providing a visual representation of the cause and effect of disaster sub-events.