融合词向量的多特征问句相似度计算方法研究  

Research on Question Similarity Computation Method of Multi-Feature with Word Embedding

作　　者：曹莉丽[1] 王未央[1] 

机构地区：[1]上海海事大学信息工程学院,上海201306

出　　处：《现代计算机（中旬刊）》2017年第6期55-59,共5页Modern Computer

摘　　要：问句的相似度计算是自动问答系统的关键部分,而现有的计算模型准确率较低。为此,应用Word2vec建立带有语义的词的向量表示,提出一种利用词向量距离,结合词项权重、句子长度等因素的问句相似度计算方法。该方法综合考察两个句子的语义信息,句法信息,词的表层信息来测量问句的相似度。实验部分设计四种问句相似度计算方法,结果表明该方法提高相似度计算的准确率。Question similarity calculation is the key of the automatic question answering system. Now, the accuracy of question similarity calculation is low. Therefore, presents a question similarity model that takes into account word weight, question length and other factors with semantic word vectors being established using Word2vec. This method comprehensively considers the semantic information, syntactic information and word information of two sentences to calculate the similarity of question sentences. Finally, designs four different question similarity calculation methods, the experimental results show that the method improves the accuracy of question similarity calculation.

关 键 词：词向量 依存关系 问句相似度 词项权重 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]