基于信息抽取技术的问答系统  

Question answering system based on information extraction techniques

作　　者：于根 李晓戈[1] 刘睿[1] 范贤[1] 杜丽萍[1] 

机构地区：[1]西安邮电大学计算机学院,陕西西安710121

出　　处：《计算机工程与设计》2017年第4期1051-1055,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(61373116);陕西省普通高等学校重点学科专项资金建设基金项目(112-1602);西安邮电大学研究生创新基金项目(ZL2013-34)

摘　　要：通过分析实体关系和命名实体,提出基于层次的答案提取方法。在将问题分为实体关系型、实体型和关键词型3类的基础上,按照实体关系层、实体层、关键词层得到答案集,利用基础特征、命名实体匹配和实体关系匹配进行特征排序提取答案。基于NLPCCEVAL2015的测试结果表明,在考虑命名实体和实体关系的情况下,准确率比仅使用基础特征的情形提高了6.1%,达到54.05%。After analyzing the entity relation and the named entity, a hierarchical method was proposed. The problem was divided into entity relation type, named entity type and keyword type. The answer set was collected according to the entity relation, the entity and the keyword, and the final answer was obtained by ranked features that included basic features, matching named entity and matching entity relation. Experimental results based on the NLPCCEVAL2015 show that the accuracy rate reaches 54. 05% when using named entities and entity relations, the result is increased by 6.1% compared to that considering basic features only.

关 键 词：问答系统 信息抽取 实体关系 命名实体 层次法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]