基于知识图谱的军事武器问答系统  

Military Weapon QA System Based on Knowledge Graph

作　　者：李代祎 盛杰 刘运星 马宗民 LI Daiyi;SHENG Jie;LIU Yunxing;MA Zongmin(School of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,南京211106

出　　处：《指挥信息系统与技术》2020年第5期58-65,共8页Command Information System and Technology

基　　金：军委科技委国防科技创新特区资助项目。

摘　　要：以现有武器装备关系型数据和非结构化军事网页数据为基础构建了一种基于知识图谱的军事武器智能问答系统。首先,针对问答系统中的问题理解,构建支持向量机(SVM)多分类器进行问句分类;接着,采用双向长短时记忆网络(BiLSTM)和条件随机场(CRF)完成命名实体识别任务,理解问句中的实体、属性和关系;然后,将用户问题匹配到最相似的问句模板上;最后,使用生成的图数据库查询语句Cypher从已构建的知识图谱中查询答案。试验结果表明,该问答系统在军事武器领域应用中具有较高准确率和使用价值。Based on the existing structured weapon equipment relational data and the semi-structured military web page data,an intelligent question answering(QA)system in the field of military weapons based on the knowledge graph(KG)is constructed.Firstly,aimed at the understanding of questions in QA system,a multi classifier based on support vector machine(SVM)is constructed for the question classification.Secondly,the bi-directional long short-term memory(BiLSTM)and conditional random field(CRF)are combined to complete the task of named entity recognition,and to understand the entities,attributes and relationships in question sentences.Then,the user's question is matched with the most similar question template.Finally,the generated query statement Cypher in the graph database is used to query the answer from the constructed knowledge graph.Experimental results show that the QA system has high accuracy and application value in the application of the field of military weapons.

关 键 词：军事武器 知识图谱 知识问答 支持向量机 问句解析 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]