摘要
针对军事实体识别依赖领域知识的特点,提出了融合本体特征的BiLSTM-CRF军事实体识别模型。通过构建军事领域本体,将分词的本体特征作为领域知识融入到词向量中,有效弥补了传统命名实体识别方法的领域知识缺乏问题,且模型加入字向量描述分词的内部形态学特征,避免了分词不准确与未登录词对军事命名实体识别造成的影响,最后采用BiLSTM-CRF模型实现军事命名实体识别工作。实验证明,该军事实体识别模型的准确率达到91.08%,能够有效识别军事实体。
According to the characteristics of military entity recognition that depends on domain knowledge,Military entity recognition model was proposed combing ontology and Bi-LSTM-CRF(Bidirectional Long Short Term Memory with Conditional Random Field).Through the domain ontology construction of military equipment,it effectively makes up for the lack of domain knowledge in traditional named entity recognition methods.Word vectors are added to the model to describe the internal morphological features of the participle,which makes up for the lack of domain knowledge effectively.And vector method describes the morphological features of the interior of the participle avoiding the unknown words from inaccurate participle and the effects on the naval ordnance named entity recognition.Finally the model was realized based on BiLSTM-CRF.The experiment proves that the identification accuracy of military named entity identification method combining ontology and Bi-LSTM-CRF is 91.08%.It can effectively solve the problem of military named entity identification.
作者
齐玉东
丁海强
吴晋豫
司维超
QI Yudong;DING Haiqiang;WU Jinyu;SI Weichao(Naval aviation university,Yantai 264001,China;The No.92199 th Troop of PLA,Qingdao 266000,China)
出处
《兵器装备工程学报》
CAS
北大核心
2020年第5期118-123,共6页
Journal of Ordnance Equipment Engineering
基金
山东省重点研发计划项目(2016YJS02A01)。
