摘要
为提取灾害性新闻中的基本要素,迅速掌握灾害事件信息和发展趋势,将目标分类和命名实体识别(named entity recognition,NER)相结合,提出改进的灾害新闻3要素提取方法。构建滑动窗检测器搭载不同的分类模型,实现对新闻文本的灾害主题识别与时空位置要素范围判定,结合命名实体识别完成对时空位置要素的精准提取,并以灾害事故信息文本为例进行测试。研究结果表明:通过在火灾、地震和滑坡新闻中进行数据集中测试,发现本文方法相较于LSTM,BILSTM,BILSTM-CRF提取效果更优;本文方法可对大量灾害性新闻的灾害3要素进行识别提取,对灾害信息进行时空规律分析,研究结果可在灾害应急响应中发挥重要作用。
In order to quickly grasp the information and development trend of disaster events,combined the extraction of basic elements of disaster news with the research of object classification and named entity recognition(NER),an improved extraction method for three elements of disaster news was proposed.Asliding window detector was constructed to carry different classification models,which achieved to recognize the disaster topic and determine the scope of spatio-temporal location elements of the news text,and then completed the accurate extraction of spatio-temporal location elements combined with NER.The test was carried out by taking the disaster accident information text as an example.The results showed that compared with LSTM,BILSTM and BILSTM-CRF methods,the F1 value of the proposed method increased by 18.6%,7.5%and 0.8%,respectively.The method could quickly recognize and extract the disaster three elements of a large number of disastrous news,so as to analyze the spatio-temporal law of disaster information.The research results can play an important role in the disaster emergency response.
作者
牛飞
钟少波
刘楠
钟伟齐
杨德威
叶欣澜
梅新
NIU Fei;ZHONG Shaobo;LIU Nan;ZHONG Weiqi;YANG Dewei;YE Xinlan;MEI Xin(Faculty of Resources and Environment,Hubei University,Wuhan Hubei 430062,China;Institute of Urban Systems Engineering,Beijing Academy of Science and Technology,Beijing 100035,China;College of Resource Environment and Tourism,Capital Normal University,Beijing 100048,China)
出处
《中国安全生产科学技术》
CAS
CSCD
北大核心
2023年第2期13-19,共7页
Journal of Safety Science and Technology
基金
国家自然科学基金项目(72174031)
北京市科学技术研究院北科青年学者计划项目(YS202004)。
关键词
灾害新闻
灾害新闻3要素
自然语言处理
目标分类
命名实体识别
disaster news
three elements of disaster news
natural language processing(NLP)
object classification
named entity recognition(NER)
