Scholarlink

This research addresses the critical gap in bushfire evacuation studies: the lack of empirical data on human behavior during evacuations. Social media platforms offer a rich source of data, but effectively mining and analyzing this data presents significant challenges due to its unstructured nature, noise, ambiguity and large volume. To overcome these challenges, we propose Queries-Optimized LLM-Empowered Active Learning (QOLEAL), a novel data mining approach that leverages a large language model (LLM) and active learning techniques. QOLEAL accurately filters and refines social media posts, identifying relevant information about evacuation decisions, destinations, and locations. The LLM’s efficiency is significantly improved through batched-operation optimization. QOLEAL demonstrates superior accuracy compared to existing approaches, achieving high precision and F1 scores while maintaining comparable recall. The method’s effectiveness is validated through experiments on three real-world datasets. Furthermore, ablation tests highlight the positive impact of batch size optimization on runtime efficiency.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Queries Optimised LLM-Empowered Active Learning for Social Media Analysis of Human Behaviour in Bushfire Evacuations

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Junfeng Wu,
  • Xiangmin Zhou,
  • Erica Kuligowski,
  • Yanchun Zhang

摘要

This research addresses the critical gap in bushfire evacuation studies: the lack of empirical data on human behavior during evacuations. Social media platforms offer a rich source of data, but effectively mining and analyzing this data presents significant challenges due to its unstructured nature, noise, ambiguity and large volume. To overcome these challenges, we propose Queries-Optimized LLM-Empowered Active Learning (QOLEAL), a novel data mining approach that leverages a large language model (LLM) and active learning techniques. QOLEAL accurately filters and refines social media posts, identifying relevant information about evacuation decisions, destinations, and locations. The LLM’s efficiency is significantly improved through batched-operation optimization. QOLEAL demonstrates superior accuracy compared to existing approaches, achieving high precision and F1 scores while maintaining comparable recall. The method’s effectiveness is validated through experiments on three real-world datasets. Furthermore, ablation tests highlight the positive impact of batch size optimization on runtime efficiency.