基于多特征的汉语句子相似度计算模型的研究

Research on Chinese Sentence Similarity Calculation Model Based on Multi-features

句子相似度的计算在自然语言处理的各个领域中都占有很重要的地位。文中深入分析了现有的一些句子相似度计算的方法,这些方法各自从词特征、词义特征或句法特征等某一侧面描述了句子相似的情况,未能全面地描述一个句子的完整信息。文中提出了一种新的基于多特征的汉语句子相似度的计算模型。该方法在基于词的基础上,从句子中词的表层到词的逻辑联系,从句子的局部结构到整体结构,用句子的区分度、相同词的相似度、长度相似度、词性相似度及词序相似度五个方面来综合考虑两个句子相似度的计算。实验结果表明,该方法合理、简便、可行。

Sentence similarity calculation plays an important role in various areas of natural language processing.Analyze the existing some sentence similarity calculation method.These methods describe the sentence similarity from the word characteristics,semantic features or syntactic features,all the information of a sentence can＇t be described fully.A new model of Chinese sentence similarity based on the multi-feature is proposed.This method is based on the word,from the surface to the logical connection of the word,from local structure to the overall structure of a sentence,five aspects of sentence similarity such as degree of differentiation,the same word similarity,length similarity,the part of speech similarity and word order similarity have been studied in depth.Experimental results show that the method is reasonable,simple and feasible.