基于构式语块的句子相似度计算

针对当前主要的汉语句子相似度计算方法存在的问题,通过对大量的中文语句进行构式语义分析,提取构式语块特征,构建构式义特征库,并提出了一种基于构式语块的中文句子相似度计算方法.该方法最大化语块粒度,缩小语义理解间隙.在计算时,首先,通过自上而下语块分析方法,利用HMM学习算法,对复杂多样的汉语句子进行构式语块识别；然后,采用自底向上的计算方式,获取得到构式块间的相似度；最后,计算得出句子间的相似度量值.对比句子相似度实验结果表明,论文提出的计算方法优于其他算法.

Method for Chinese Sentence Similarity Computation Based on Construction-chunk

Aiming at the shortcomings of current similarity computation methods of Chinese sentence,through analyzing construction meaning on a mass of Chinese sentences,features of construction-chunk are retrieved,and feature base of construction meanings is built.Then a similarity computation method of Chinese sentence based on construction-chunk is proposed.This approach maximizes chunk granularity and narrows the gap between semantic comprehensions.Using this method,construction-chunk of complex and diversified Chinese sentences are recognized by up-to-down construction-chunk parsing theory and HMM learning method.By down-to-up calculation method,the similarity of each construction-chunk in the two sentences is computed separately,and then the similarity computation of sentence is realized,which is based on construction-chunk similarity.Finally,the experimental result shows that the proposed method has better performance than the others do.