Optimizing Emergency Decision-Making for Natural Disaster Using Light Retrieval Augmented Generation

Emergency decision support system can enhance the disaster management efficiency by generating scientific and real-time emergency plans. It has developed increasingly recent years, especially with the integration of large language model (LLM). However, Emergency decision-making in natural disaster area relies on regional-specific knowledge, thus in this paper we utilize the light retrieval augmented generation (RAG) with LLM for natural disaster area. We further apply few-shot prompting and chain-of-thought fine-tuning technique to optimize the system. The experiments prove the effectiveness of light rag for decision generation, which can significant improve the performance on our corpus.