基于Word2vec和句法规则的自动问答系统问句相似度研究  

QUESTION SIMILARITY OF AUTOMATIC QUESTION ANSWERING SYSTEM BASED ON WORD2VEC AND SYNTACTIC RULES

作　　者：刘杰 白尚旺[1] 陆望东 党伟超[1] 潘理虎[1,3] Liu Jie;Bai Shangwang;Lu Wangdong;Dang Weichao;Pan Lihu(School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan 030024,Shanxi,China;Taiyuan Zhengtongyun Technology Co.,Ltd.,Taiyuan 030000,Shanxi,China;Institute of Geographic Science and Natural Resource Research,Chinese Academy of Science,Beijing 100101,China)

机构地区：[1]太原科技大学计算机科学与技术学院,山西太原030024 [2]太原政通云科技有限公司,山西太原030000 [3]中国科学院地理科学与资源研究所,北京100101

出　　处：《计算机应用与软件》2021年第3期169-174,201,共7页Computer Applications and Software

基　　金：山西省中科院科技合作项目(20141101001);山西省重点研发计划(一般)工业项目(201703D121042-1);山西省社会发展科技项目(20140313020-1)。

摘　　要：自动问答系统问句相似度计算的准确率直接影响系统返回答案的准确率,对此提出一种基于Word2vec和句法规则的问句相似度计算方法。构造Text-CNN问句分类模型将问句进行分类,再构造Word2vec词向量模型将问句中词与词的空间向量相似度转换成语义相似度,并加入句法规则的分析。随机从搜狗公开问答数据集中抽取200条数据进行测试,结果表明,该方法与TF-IDF方法相比,自动问答系统返回答案的准确率和召回率分别提高了0.259和0.154。The accuracy of question similarity calculation in the automatic question answering system directly affects the accuracy of the answers returned by the system.Therefore,a question similarity calculation method based on Word2vec and syntactic rules is proposed.This method constructed the Text-CNN questions classification model to classify questions,and then constructed the Word2vec word vector model to convert the spatial vector similarity of words and words in questions into semantic similarity,and added the analysis of syntactic rules.The 200 data were randomly extracted from Sogou public Q&A data set for testing.The results show that compared with TF-IDF method,the accuracy rate and recall rate of the automatic question answering system are improved by 0.259 and 0.154 respectively.

关 键 词：自动问答系统 Word2vec Text-CNN 问句相似度 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]