一种基于卷积神经网络的自动问答系统  

A Question Answering System Based on Convolutional Neural Networks

作　　者：杨骏[1,2] 敬思远[1,2] 项炜[1,2] YANG Jun;JING Siyuan;XIANG Wei(School of Computer Science,Leshan Normal University,Leshan Sichuan 614000,China;Key Lab of Internet Natural Language Processing of Sichuan Provincial Education Department,Leshan Normal University,Leshan Sichuan 614000,China)

机构地区：[1]乐山师范学院计算机科学学院,四川乐山614000 [2]互联网自然语言智能处理四川省高等学校重点实验室(乐山师范学院),四川乐山614000

出　　处：《乐山师范学院学报》2020年第4期40-44,共5页Journal of Leshan Normal University

基　　金：乐山市科技局重点研究项目“‘乐山美食’”机器问答系统的关键技术研究”(18JZD091)和“基于深度神经网络的汉语语用错误检测研究”(18JZD117).

摘　　要：文章基于卷积神经网络等技术构建了一个自动问答系统。该系统使用jieba分词组件进行中文分词,使用人工智能标记语言(AIML)进行基于规则的问题匹配;对于规则无法匹配的问题,通过常用的搜索引擎进行文本检索或知识图谱检索,对检索得到的多个答案使用卷积神经网络进行语义相似度计算,根据相似度对答案进行排序,最后筛选出最可信的答案返回给用户。研究表明,系统响应延时虽然还不够理想,但已经具备了初步应用的条件。A question answering system based on convolutional neural networks is developed. In the system,jieba is used for Chinese word segmentation,and AIML is used for matching a question with the assigned rules. To those questions cannot be matched with these rules,the search engines are used to get the answers based on text or knowledge graph,and then the answers’ ranking is determined by the semantic similarities computed by CNN between the question and the answers. At last, the best answer is returned to the user. Experiments indicate that the system can be practicable,even though its response delay is not perfect.

关 键 词：问答系统 卷积神经网络 AIML 中文分词 

分 类 号：TP311.52[自动化与计算机技术—计算机软件与理论]