摘要
从大量非结构化的作战文书中抽取关键事件信息并以结构化方式描述,是构建作战知识图谱、挖掘隐性战场信息、动态更新战场态势图的基础工作。我军体制编制调整改革常态化运行后,作战文书中涌现出诸多新型作战力量、新式部队番号和新颖作战运用方法,依靠人工构建模板提取作战文书事件的传统方法已难以为继,重新构建模板不仅异常复杂,而且通用性和泛化性不强。针对以上问题,提出一种基于深度学习的作战文书事件抽取方法。结合双向长短时记忆(Bi-directional Long Short-Term Memory,Bi-LSTM)神经网络对较长句子上下文的记忆能力、动态字向量(Embedding from Language Models of Character,ELMo)对汉字语义的多重表示能力和条件随机场(Conditional Random Field,CRF)对标注规则的学习能力,构建了基于ELMo+Bi-LSTM+CRF作战文书事件抽取模型。为验证方法的有效性,在演习导调文书语料集上进行了实验,实验结果表明,该方法抽取效果较好,对于从大规模作战文书文本中抽取事件信息来说,能够满足一定的应用需求。
Extracting key event information from a large number of unstructured operational documents and describing them in a structured form is very important,which is the basic work of constructing operational knowledge graph,mining hidden battlefield information,and dynamically updating maps of battlefield situation.After the normalization of our military system and establishment adjustment and reform,many new operational forces,new unit designation and new tactics have emerged in the operational documents.The traditional method of manually constructing templates to extract operational documents events has been unsustainable,since rebuilding the template is not only extremely complicated,but also not very versatile and generalized.In view of the above problems,this paper proposes a method of operational documents event extraction based on deep learning.Combining the ability of bi-directional long short-term memory(Bi-LSTM)neural network to learn the long sentence context and the multiple representation of Chinese characters in the embedding from language models of character(ELMo)and the conditional random field(CRF)learning ability of the tagging rules,the ELMo+Bi-LSTM+CRF operational documents event extraction model is constructed.In order to verify the validity of the method,experiments are carried out on the corpus of the exercise guidance documents.The experimental results show that the method has a good extraction effect and can meet certain application requirements for extracting event information from the text of large-scale operational documents.
作者
王学锋
杨若鹏
李雯
WANG Xuefeng;YANG Ruopeng;LI Wen(Institute of Communications, National Uni. of Defense Tech, Wuhan 430010, China)
出处
《信息工程大学学报》
2019年第5期635-640,共6页
Journal of Information Engineering University
基金
军队科研基金资助项目。
关键词
作战文书
事件抽取
深度学习
operational documents
event extraction
deep learning
