面向自动问答系统的问句相似度计算研究

Question similarity computing method for automatic question answering system

结构化自动问答系统采用传统方法缺少对词汇、词序和结构的划分,导致语句相似度较低,为了解决该问题,提出了基于Web语义的混合问句相似度计算方法。根据结构化自动问答系统结构,设计系统语句分析模型,通过正向匹配方法,对模型专业词库中的用户输入自然语句进行分词处理,并对字符串之间的关系展开分析。采用非恒定相似度系数来描述2个字符串的相似情况,并由此分析词形、词序和结构相似度,完成不同语句相似度的计算。通过实验对比可知,文章提出的基于Web语义的混合问句相似度计算方法最高计算精准度可达到96%,可提升自动问答系统的整体性能。

Structured automatic question answering system lacks the division of vocabulary,word order and structure,which leads to low sentence similarity.In order to solve this problem,a method of computing sentence similarity based on Web semantics is proposed.According to the structure of the structured automatic question answering system,the system statement analysis model is designed.Through the forward matching method,the user input natural statements in the model professional lexicon are segmented and the string relationship is analyzed.The unsteady similarity coefficient is used to describe the similarity of two strings,and the similarity of morphology,word order and structure is analyzed to determine the similarity of different sentences.The experimental results show that with this method,the minimum computational accuracy is 42%,and the maximum computational accuracy can reach 96%.The overall computational accuracy has been greatly improved compared with that based on neural network and ant colony,thereby improving the overall performance of the automatic question answering system.