📄 中文摘要
及时分类社交媒体上的人道信息对有效的灾害响应至关重要。然而，在资源有限的紧急环境中部署大型语言模型（LLMs）面临挑战。该研究开发了一种轻量级、成本效益高的框架，用于灾害推文分类，采用参数高效的微调方法。通过整合和规范化HumAID数据集（包含76,484条推文，涵盖19个灾害事件），构建了一个统一的实验语料库，形成了人道信息分类和事件类型识别的双任务基准。通过对提示策略、LoRA微调和检索增强生成（RAG）在Llama 3.1 8B上的系统评估，验证了该框架的有效性。
灾害人道信息分类的轻量级大语言模型框架
📄 English Summary
A Lightweight LLM Framework for Disaster Humanitarian Information Classification
Timely classification of humanitarian information from social media is crucial for effective disaster response. However, deploying large language models (LLMs) in resource-constrained emergency settings presents significant challenges. This study develops a lightweight, cost-effective framework for disaster tweet classification using parameter-efficient fine-tuning. A unified experimental corpus is constructed by integrating and normalizing the HumAID dataset, which contains 76,484 tweets across 19 disaster events, into a dual-task benchmark for humanitarian information categorization and event type identification. Systematic evaluation of prompting strategies, LoRA fine-tuning, and retrieval-augmented generation (RAG) on Llama 3.1 8B demonstrates the effectiveness of the proposed framework.