Predicting Human Mobility in Disasters via LLM-Enhanced Cross-City Learning

2025-07-25

AI Summary

Rapid urbanization and climate change have heightened urban vulnerability to natural disasters, yet existing mobility prediction models fail to capture anomalous human movement patterns during emergencies, impeding pre-disaster early warning and resource prepositioning. To address this, we propose DisasterMobLLM—a novel framework that pioneers the integration of large language models (LLMs) into disaster-aware mobility forecasting. It comprises three core components: (1) an RAG-enhanced intent predictor modeling individual emergency mobility motivations; (2) an LLM-based intent refinement module enabling cross-city transfer of disaster response knowledge; and (3) an intent-modulated location predictor improving modeling fidelity for non-routine behavior. Extensive experiments demonstrate that DisasterMobLLM achieves a 32.8% improvement in Acc@1 and a 35.0% gain in F1-score for stationary-state prediction over state-of-the-art baselines, significantly advancing robust, context-aware urban mobility forecasting under crisis conditions.

Technology Category

• Planning, Routing, and Scheduling: Planning with Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The vulnerability of cities to natural disasters has increased with urbanization and climate change, making it more important to predict human mobility in the disaster scenarios for downstream tasks including location-based early disaster warning and pre-allocating rescue resources, etc. However, existing human mobility prediction models are mainly designed for normal scenarios, and fail to adapt to disaster scenarios due to the shift of human mobility patterns under disaster. To address this issue, we introduce extbf{DisasterMobLLM}, a mobility prediction framework for disaster scenarios that can be integrated into existing deep mobility prediction methods by leveraging LLMs to model the mobility intention and transferring the common knowledge of how different disasters affect mobility intentions between cities. This framework utilizes a RAG-Enhanced Intention Predictor to forecast the next intention, refines it with an LLM-based Intention Refiner, and then maps the intention to an exact location using an Intention-Modulated Location Predictor. Extensive experiments illustrate that DisasterMobLLM can achieve a 32.8% improvement in terms of Acc@1 and a 35.0% improvement in terms of the F1-score of predicting immobility compared to the baselines. The code is available at https://github.com/tsinghua-fib-lab/DisasterMobLLM.

Problem

• Predict human mobility in disaster scenarios
• Enhance cross-city learning via LLMs
• Improve disaster response with accurate mobility forecasts

Innovation

• LLM-enhanced cross-city learning for mobility prediction
• RAG-enhanced intention predictor for disaster scenarios
• Intention-modulated location predictor for exact mapping