Leveraging Social Media Data for Disaster Management in the Era of Large Language Models

In the event of crises, social media provides comprehensive situational knowledge that can facilitate disaster management and response. However, extracting meaningful insights from social media data is a complex task. To this end, the literature justifies the unprecedented use of large language models (LLMs) for natural language and in-depth understanding. Researchers have extensively explored the use of LLMs in disaster management and response to mitigate the effects and support distressed victims. In this study, we comprehensively investigate the capabilities of LLMs to perform a multifaceted analysis of social media use for crisis mitigation. Using a structured systematic review methodology, this paper reviewed 120 studies emphasizing: (1) the importance of social media data and the integration of LLMs into disaster relief programs, (2) the use of LLMs for sentiment and emotion detection to understand the psychological impact of disasters from social media posts, (3) the challenges in the responsive utilisation of social media data for decision-sensitive applications that impact at-risk lives. Furthermore, we focus on critical areas of disaster management and response in which LLMs warrant exploration. We inspect the opportunities, limitations, ethical concerns, and risks associated with the identified areas. The review provides a comprehensive guide for researchers and disaster management organizations to draw useful insights that strengthen their understanding of disaster management and response. Its applications can be used to design social media-based emergency response systems that integrate multimodalities and LLMs to enhance situational awareness for disaster restoration.