基于概念向量的文本语义相似度方法探索

摘要

Based on the previous studies on the concept semantic similarity, this paper proposed measurement of text semantic similarity on the basis of concept vector. First, mining the concepts or terms from the texts. Second, transforming concepts or terms into concept vector followed by hierarchical structure of vocabulary. At last, measuring the sematic similarity of concepts or terms and further measuring the text semantic similarity. The paper used TREC-05 genomics track data to experiment. The results showed that the method of text semantic similarity on the basis of concept vector was bet er than cosine, which was more closely to expert evaluation result.%在对概念语义相似度方法调研的基础上,本文提出基于概念向量的文本语义相似度测度方法,借助MetaMap工具抽取文本中的概念术语,将概念术语通过词表层级结构转化为概念向量,通过计算两文本中概念向量的语义相似度来测度两文本的语义相似度.为验证基于概念向量文本语义相似度方法的准确性,选取TREC-05 genomics track数据进行实验,实验结果表明,本文提出的方法较常用的余弦方法更优,与专家评估方法更接近,在测度文本语义相似度上具有一定的可行性和有效性.