A Large Language Model‐Driven Framework for Earthquake Emergency Mapping and Analysis From Social Media Text

ABSTRACT

Timely and reliable spatiotemporal information is critical for effective post‐earthquake emergency response. However, social media data are often noisy and unstructured, making it challenging to derive actionable information from rapidly generated content. This study develops a GIScience‐oriented framework that uses large language models for earthquake emergency mapping and analysis. The framework integrates an earthquake emergency mapping ontology with instruction‐tuned LLMs to perform multi‐label classification, fine‐grained attribute extraction, and structured representation of disaster information. Experimental results show that the framework achieves reliable performance in identifying Disaster Situation , Rescue Information , and Help Request , while accurately extracting key spatial attributes such as location. The case study of the 2023 Jishishan earthquake further reveals spatial relationships among disaster impacts, assistance demands, and emergency responses. The framework converts heterogeneous social media texts into structured geospatial evidence, providing support for situational awareness and decision‐making during earthquake emergencies.