CMNEE(Chinese Military News Event Extraction dataset)
收藏资源简介:
CMNEE(Chinese Military News Event Extraction dataset)是国防科技大学、东南大学和清华大学联合构建的一个大规模的、基于文档标注的开源中文军事新闻事件抽取数据集。该数据集包含17,000份文档和29,223个事件,所有事件均基于预定义的军事领域模式人工标注,包括8种事件类型和11种论元角色。数据集构建遵循两阶段多轮次标注策略,首先通过权威网站获取军事新闻文本并预处理,然后依据触发词字典进行预标注,经领域专家审核后形成事件模式。随后,通过人工分批、迭代标注并持续修正,直至满足既定质量标准。CMNEE作为首个专注于军事领域文档级事件抽取的数据集,对推动相关研究具有显著意义。
CMNEE (Chinese Military News Event Extraction dataset) is a large-scale, document-annotated open-source Chinese military news event extraction dataset jointly constructed by the National University of Defense Technology, Southeast University and Tsinghua University. The dataset contains 17,000 documents and 29,223 events, all of which are manually annotated based on pre-defined military domain schemas, including 8 event types and 11 argument roles. The dataset construction follows a two-stage, multi-round annotation strategy: first, military news texts are obtained from authoritative websites and preprocessed, then pre-annotation is conducted via a trigger word dictionary, and the event schemas are finalized after review by domain experts. Subsequently, batch-by-batch, iterative manual annotation and continuous correction are carried out until the established quality standards are met. As the first document-level event extraction dataset focused on the military domain, CMNEE holds significant importance for promoting relevant research.
CMNEE数据集概述
数据集基本信息
- 名称: CMNEE
- 类型: 大规模文档级中文军事新闻事件抽取数据集
- 文档数量: 17,000
- 事件数量: 29,223
- 事件类型: 8种
- 论元角色类型: 11种
- 数据来源: 开源中文军事新闻
- 数据获取: Google Drive链接
数据集结构
- 数据格式: JSON
- 核心字段:
id: 实例唯一标识text: 包含事件信息的文档文本event_list: 标注的事件列表,每个事件包含event_type,trigger(触发词),arguments(论元)coref_arguments: 实例中的共指论元列表,每个子列表指向文本中的同一对象
数据集应用
- 相关竞赛: 用于支持“特定领域多事件信息联合抽取”竞赛,竞赛平台为DataFountain
- 竞赛特点:
- 数据分割: CMNEE分为训练集、验证集和测试集
- 测试标签: 测试集无标签信息
- 评估方法: 使用Precision, Recall和F1-score进行评估
数据集评估
- 基准模型:
- DCFEE-O & DCFEE-M
- GreedyDec & Doc2EDAG
- DEPPN
- BERT+CRF
- EEQA
- TEXT2EVENT
- PAIE
- 数据处理: 相关处理数据可从Google Drive链接获取
引用信息
- 论文: CMNEE: A Large-Scale Document-Level Event Extraction Dataset based on Open-Source Chinese Military News
- 作者: Zhu, Mengna et al.
- 出版: Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)




