【作者】 赵兴； 余正涛； 邹俊杰； 郭剑毅； 毛存礼；

【Author】 ZHAO Xing~(1,2),YU Zheng-Tao~(1,2),ZOU Jun-Jie~(1,2),GUO Jian-Yi~(1,2),MAO Cun-Li~(1,2) 1.School of Information and Automation,Kunming University of Science and Technology,Kunming 650051,P.R.China 2.Laboratory of Intelligent Information Processing,Kunming University of Science and Technology,Kunming 650051,P.R.China

【机构】 昆明理工大学信息工程与自动化学院； 昆明理工大学智能信息处理重点实验室；

【摘要】 针对受限领域问答系统答案句子模式的特点,提出了一种结合依存句法与领域实体的答案句子模式学习方法,该方法首先通过已标记的问题类型及答案的问题检索召回候选答案片段,对答案候选片段进行领域实体识别和依存句法分析,提取候选答案句子模式,然后通过对同类型实体进行扩展,获得新的候选答案句子模式,最后通过候选模式检索召回答案的频度和精度计算候选模式的权重,获得候选答案句法模式库。对提取的模式通过模式匹配方式进行答案抽取实验,在模式获取数据集和Web数据集上的实验表明,提出的模式学习方法能有效的改善受限领域问答系统的答案抽取效果。更多还原

【Abstract】 For characteristics of the sentence patterns in restricted domain question- answer system,we proposed an answer syntax pattern learning method combined with dependency syntax and sentence patterns of domain answers entity.Firstly,this method uses labeled type of questions and answers to recall candidate snippets,then dependency parsing and entity recognition are conducted on the snippets to extract the candidate answer sentence patterns.Next,extend the same type of entities to obtain new candidate answer sentence patterns.Finally,get the candidate answer syntax pattern corpus according to the weight of candidate pattern calculated by the frequency and precision of the answers,which retrieved by the candidate patterns.We experimented on the answers extracting by pattern matching,the experimental on web data sets and the accessed pattern data sets show that the pattern learning method could effectively improve the results of extracting answers in the restricted domain.更多还原