领域问答系统答案提取方法研究  

Research on the Method of Answer Extraction in Domain Chinese Question Answering System

作　　者：毛存礼[1,2,3] 余正涛[1,2] 韩露[1] 郭剑毅[1,2] 李丽娜[1] 雷雄丽[4] 

机构地区：[1]昆明理工大学信息工程与自动化学院,云南昆明650051 [2]云南省计算机技术应用重点实验室智能信息处理研究所,云南昆明650051 [3]昆明冶金高等专科学校计算机学院,云南昆明650093 [4]昆明冶金高等专科学校建筑工程与艺术学院,云南昆明650093

出　　处：《烟台大学学报（自然科学与工程版）》2009年第3期212-216,共5页Journal of Yantai University(Natural Science and Engineering Edition)

基　　金：国家自然科学基金资助项目(60863011);云南省自然科学基金重点资助项目(2008CC023);云南省中青年学术和技术带头人后备人才基金资助项目(2007PY01-11);云南省教育厅基金重点资助项目(07Z11139);云南省教育厅基金资助项目(08C0220)

摘　　要：在特定领域问答系统中,领域知识直接影响问答效果.本文提出了一种领域问答答案提取方法,以问题分析得到问题查询、问题类型及答案类型为基础,借助领域知识检索获得答案提取候选段落.对于定义性问题,结合关键词加权权重计算方法及句子与问题语义相似度方法,对候选段落或句子与问题相关度排序,提取相关度高的句子或段落作为答案,对于数词或列表性实体问题,借助命名实体识别,提取与问题中心相关的领域实体作为答案.在云南旅游领域进行了答案提取实验,结果表明该方法具有较好的效果.The domain knowledge has a direct influence on the result of question-and-answer in the specific domain question answering system. In this paper, a method of answer extraction from domain question-and-answer is presented, in which candidate paragraphs can be obtained for answer extraction with the help of domain knowledge retrieval based on question query, question types and answer types. For definition questions, sentences or paragraphs with higher relevance can be extracted to become the answer based on the relevance ranking between questions and the candidate sentences or paragraphs by combining the computing method of keywords weighting with the method of semantic similarity between the sentences and questions. For questions of numeral or list entity type, with the help of named entity recognition, the domain entity related to the key part of the questions is extracted to be the answer. The experiments on the answer extraction have been carried out in the field of Yunnan tourism and the results show that favorable effects have been achieved by using the method of answer extraction from domain question-and-answer.

关 键 词：问答系统 领域知识 问题类型 答案类型 答案提取 

分 类 号：TP391.12[自动化与计算机技术—计算机应用技术]