摘要
文中主要研究了自动问答系统的句子相似度的几种常见算法,基于统计的VSM算法、语义相似度算法、结构的相似度算法,并在此基础上进行改进,提出了一种新的句子相似度算法,提高了自动问答系统的查全率和查准率。主要研究了分词、标注词性和权值、计算词语的相似度,进而计算句子的加权相似度,最后从FAQ中抽取相似度比较高的句子以及答案给用户。最后通过实验进行验证,这种新的句子相似度算法在一定程度上提高了自动问答系统的查全率和查准率,并具有一定的合理性。
Mainly study on several sentence similarity approaches of the automatic ask and answer system, such as VSM algorithm based on statistics, semantic similarity algorithm and structure similarity. Based on this, take weight into account to improve the similarity and put forward the new sentence similarity method,in order to improve the recall and precision of ask ＆ answer system. It mainly focuses on the segmentation, and markes the characteristics and weight of words, removing the stop words, computing the semantic similarity between two words based on the hownet,computing the weighted similarity between two questions, choose the more similar questions and answers from FAQ hack to the users at last. Experiment results show the algorithm improves the recall and precision at some extent.
出处
《计算机技术与发展》
2012年第5期75-78,共4页
Computer Technology and Development
基金
中国科学院计算技术研究所国家重点实验室开放课题(2009JS095J)
关键词
分词
本体
知网
语义相似度
查全率
查准率
segmentation
ontology
hownet
semantic similarity
recall
precision
