摘要
对突发事件进行标注对应急管理响应能力的提升有重要意义。针对突发事件类型繁多,难以进行有效的归纳的特点,本文提出一种基于预训练语言模型BERT的事故标注方法:BERT对无序的突发事件文本提取结构化数据进行主题建模构建数据集;BERT-DPCNN模型在后期对数据集进行突发事件标注。基于构建的数据集,BERT-DPCNN加权平均的F1值达到0.9741,优于其他对比模型。结果表明:本文提出的方法可以对突发事件进行有效标注。
The annotation of emergency incidents is important for the improvement of emergency management response capability.This paper proposes a method for emergency label based on the pre-trained language model BERT:BERT extracts structured data from unordered emergency texts to build the dataset by topic model;the BERT-DPCNN model labels the dataset with emergency events at a later stage.Based on the constructed dataset,the weighted average F1 value of BERT-DPCNN reached 0.9741,which is better than other comparative models.The results show that the method proposed in this study can provide effective annotation of emergency incidents.
作者
王德志
陈靖耀
WANG Dezhi;CHEN Jingyao(School of Computer Engineering,North China Institute of Science and Technology,Yanjiao,065201,China)
出处
《华北科技学院学报》
2021年第6期74-82,共9页
Journal of North China Institute of Science and Technology
基金
国家重点研发计划项目(2018YFC0808306)
河北省物联网监控工程技术研究中心项目(3142018055)。
关键词
文本多分类
BERT
主题模型
事故标注
text multi-classification
BERT
topic model
emergency labeling
