基于Bert_TextCNN模型的漏洞知识库问答系统  

Vulnerability knowledge base Q&A system based on Bert_TextCNN model

作　　者：应王萍 宋建华[2,3,4] YING Wangping;SONG Jianhua(School of Computer Science and Information Engineering,Hubei University,Wuhan 430062,China;School of Cyber Science and Technology,Hubei University,Wuhan 430062,China;Engineering Research Center of Hubei Province in Intelligent Government Affairs and Application of Artificial Intelligence,Wuhan 430062,China;Hubei Province Project of Key Research Institute of Humanities and Social Sciences at Universities(Research Center of Information Management for Performance Evaluation),Wuhan 430062,China)

机构地区：[1]湖北大学计算机与信息工程学院,湖北武汉430062 [2]湖北大学网络空间安全学院,湖北武汉430062 [3]智慧政务与人工智能应用湖北省工程研究中心,湖北武汉430062 [4]湖北省高校人文社科重点研究基地(绩效评价信息管理研究中心),湖北武汉430062

出　　处：《湖北大学学报（自然科学版）》2023年第6期899-907,共9页Journal of Hubei University：Natural Science

基　　金：国家自然科学基金(62102136);湖北省技术创新专项重大项目(2020AEA008);湖北省重点研发计划项目重点项目(2021BAA184、2021BAA188)资助。

摘　　要：融合知识图谱(knowledge graph)在特定领域做问答系统是当下热点之一,用户的意图识别任务是问答系统中至关重要的一步.针对不同漏洞平台管理漏洞情报存在差异、内容不全面和信息孤立,以及用户获取全面的漏洞情报困难等问题,提出构建融合CNVD和CNNVD的综合漏洞知识图谱,在此基础上构建基于Bert_TextCNN意图识别的漏洞知识库问答系统.对比5个主流模型的准确率、召回率和F1值,结果显示Bert_TextCNN模型的F1值可达96.5%,比对照组中最高的F1值高2.4%,说明在意图识别任务中Bert_TextCNN模型的意图识别能力优于其他模型.Q&A system based on Fusion Knowledge Graph in a specific domain is one of the hot spots nowadays,and the user’s intention identification task is a crucial step in the Q&A system.In response to the problems of differences in vulnerability intelligence managed by different vulnerability platforms,incomplete content and isolated information,and difficulties for users to obtain comprehensive vulnerability intelligence,we proposed to build a comprehensive vulnerability knowledge graph fused with CNVD and CNNVD,and on this basis,we built a vulnerability knowledge base Q&A system based on Bert_TextCNN intent recognition.Comparing the accuracy,recall and F1 value of five mainstream text classification models,the comparison results showed that the F1 value of Bert_TextCNN model could reach 96.5%,which was 2.4%higher than the highest F1 value of the control groups,indicating that the text classification ability of Bert_TextCNN model was better than other models in the intention recognition task.

关 键 词：漏洞库 知识图谱 意图识别 智能问答 网络安全 

分 类 号：TB391.3[一般工业技术—材料科学与工程]