摘要
为解决部分军事命名实体导致规则、统计等传统模型识别率不高的问题,提出一种基于双向长短期记忆—条件随机场(BI-LSTM-CRF)的作战文书命名实体识别方法。介绍作战文书命名实体识别的概念、特点,给出模型具体训练方法与步骤,在手工构建的数据集上进行开放性测试。结果表明,该方法能有效提升作战文书命名实体的识别准确率,模型最终的识别精确率和召回率分别达到91.40%和90.43%。
Military named entity recognition is the basis of key information extraction for combat documents.Some military named entities have complex combination and nested relationship.In addition,different services have different professional expressions.These conditions lead to a low recognition rate of traditional models based on rules and statistics.In this paper,a method of named entity recognition for combat documents based on BI-LSTM-CRF is proposed.We conduct an open test on a manually constructed dataset.The results show that our method effectively improves the recognition accuracy of named entities in combat documents.The precision and recall rate of our model are 91.40%and 90.43%.
作者
张晓海
操新文
彭双震
温玉韬
ZHANG Xiaohai;CAO Xinwen;PENG Shuangzhen;WEN Yutao(Joint Operation College,NDU of PLA,Shijiazhuang 050084,China;Hybrid Brigade of Tibet Reserve Army,Lhasa 850000,China)
出处
《信息工程大学学报》
2019年第4期502-506,512,共6页
Journal of Information Engineering University
基金
国家社会科学基金资助项目(16GJ003-051)。
