基于模式学习的中文问答系统答案抽取方法

发布时间：2018-05-28 07:55

本文选题：计算机软件 + 问答系统　；参考：《吉林大学学报(工学版)》2008年01期

【摘要】：答案抽取是中文问答系统的关键,而通常答案是借助于问题的答案句子模式抽取得到,由于答案句子模式是语言专家根据语言规则提炼获得,因此非常依赖于专家经验。针对这一局限性,提出了一种利用模式学习来获得中文答案句子模式的方法,该方法利用搜索引擎从互连网上检索相关问题文本,人工提取包含答案的句子段,并标注问题类型及答案,形成各种问题类型的问答训练语料。通过统计学习,提取候选答案句子模式,计算候选句子模式权重,并根据权重获得相应问题类型的答案句子模式。基于事实的问题答案抽取结果表明,提出的基于模式学习的方法有很好的效果,实验答案提取准确率值达到了0.28,学习方法获得的模式基本上覆盖了常规答案句子模式。
[Abstract]:Answer extraction is the key of the Chinese question answering system, but the answer is usually obtained by the answer sentence pattern of the question. Because the answer sentence pattern is extracted by the language expert according to the language rules, it is very dependent on the expert experience. Aiming at this limitation, this paper proposes a method of obtaining sentence pattern of Chinese answer by using pattern learning. The method uses search engine to retrieve the relevant question text from the Internet, and manually extracts sentence segments containing the answers. The question types and answers are marked to form question-answering training corpus of various question types. Through statistical learning, the candidate sentence pattern is extracted, the weight of candidate sentence pattern is calculated, and the answer sentence pattern of the corresponding question type is obtained according to the weight. The results of fact-based answer extraction show that the proposed model learning method has a good effect, the accuracy value of experimental answer extraction reaches 0.28, and the pattern obtained by the learning method basically covers the conventional sentence pattern.
【作者单位】：昆明理工大学信息工程与自动化学院昆明理工大学信息工程与自动化学院昆明理工大学信息工程与自动化学院昆明理工大学信息工程与自动化学院昆明理工大学信息工程与自动化学院
【基金】：国家自然科学基金项目(60663004) 高等学校博士学科点专项科研基金项目(20050007023) 云南省中青年学术和技术带头人后备人才基金项目(2007PY01-11) 云南省教育厅基金项目(07Z11139) 昆明理工大学博士基金项目(2006-12)
【分类号】：TP311.52

【参考文献】