节点文献
基于网络的自动问答系统的答案抽取方法研究
The Research on Answer Extraction Approach of Web Based Automatic Question Answeing System
【作者】 孙景广;
【作者基本信息】 沈阳航空工业学院 , 计算机应用技术, 2007, 硕士
【摘要】 自动问答系统(Question Answering System,QA)与传统的依靠关键字匹配的搜索引擎相比,能够更好地满足用户的检索需求,更准确地找出用户所需的答案,具有方便、快捷、高效等特点。 问题分类是问答系统所要处理的第一步,问题分类结果的好坏直接影响后续工作的进行。答案抽取是自动问答系统所要处理的最后一步,也是非常关键的一步。如果不能在答案抽取模块准确地抽取出正确答案,将极大地降低问答系统的性能。本文在对现有问题分类和答案抽取技术进行深入调查和研究的基础上,主要进行了以下工作: 1.为了深入理解并能从整体上把握问题的语义,得到问题的答案类型等相关信息,本文以知网为知识库,提出了基于知网的中文问题自动分类,并用最大熵模型构建了一个问题分类模型。实验结果表明,分类正确率和其他方法相比有较大的提高。 2.通过借鉴前人在答案抽取方面的相关研究成果,为了在语义理解的层面上进行答案抽取,我们以知网为知识库并利用句法分析的结果,提出了句法分析和知网相结合进行答案抽取的新方法,并重点以数字类问题进行了实验,取得了较好的结果。 3.答案验证能够对答案抽取的结果进行必要的校正和反馈,在有关研究的基础上,答案抽取的最后阶段引入了答案验证机制,并进行了初步的尝试。 4.在一个简单的问答系统上,对本文提出的方法进行了实现和测试,并对实验结果讲行了评价。
【Abstract】 Compare to conventional Keywords-based search engine, Automatic Question Answering System (QA) can satisfy the users’ retrieval command better, and it can give the answer that user need indeed, so it has the characteristics of convenience、 celerity and high efficiency.Question classification is the first task of Question Answering System, and the precision of question classification has great effect on the subsequent processes. Answer extraction is the last and very important step in QA. The precision of a QA system will drop greatly if it can not give the correct answer in answer extraction module. After a deep investigation and research at question classification and answer extraction, the main work of the paper as follows:Firstly, in order to understand the questions’ meaning better and get the information such as answer type, we propose a new Chinese question classification method—HowNet based Chinese question automatic classification, and then construct a question classification module with Maximum Entropy. The experiment results show that the correction percentage has obvious improvement than other similar experiment results.Secondly, after study the former work, in order to understand the sentences from semantic angle. We propose a new method that uses HowNet as knowledge database and combine with syntax analysis to extract answer. We experiment mainly at number questions and achieve a satisfactory result.Thirdly, answer validation can give necessary revision and feedback. After research the correlative study, we induce answer validation in the last phrase of QA.Finally, we test the method proposed in this paper by an integral QA and give the evaluation results.
【Key words】 Question Answering; HowNet; Answer Extraction; Maximum Entropy; Syntax Analysis; Answer validation;
- 【网络出版投稿人】 沈阳航空工业学院 【网络出版年期】2007年 02期
- 【分类号】TP319;TP18
- 【被引频次】13
- 【下载频次】620