rdjarbeng/who-epidemic-events
收藏资源简介:
该数据集展示了将Google的Groundsource方法(基于LLM的非结构化文本结构化提取)从洪水预测转移到流行病监测的应用。通过从3,177篇WHO疾病爆发新闻(DONs)文章中提取结构化流行病事件数据(如疾病、国家、日期、病例数、死亡数、严重程度等),并使用Qwen2.5-72B-Instruct模型进行处理。数据集包含213个训练样本,涵盖了79种独特疾病和85个国家。实验结果表明,该方法在疾病名称提取、病例数提取等方面具有较高的准确率,且非洲地区在疾病监测中得到了较好的体现。
This dataset demonstrates transferring Googles Groundsource methodology (LLM-based structured extraction from unstructured text) from flood prediction to epidemic surveillance. It extracts structured epidemic event data (e.g., disease, country, date, cases, deaths, severity) from 3,177 WHO Disease Outbreak News (DONs) articles using the Qwen2.5-72B-Instruct model. The dataset includes 213 training examples, covering 79 unique diseases and 85 countries. The results show high accuracy in disease name extraction, case count extraction, etc., and Africa is well-represented in disease surveillance.




