基于修正偏移量的句子相似度算法

Sentence Similarity Algorithm Based on Corrected Offset

在计算词语语义相似度的基础上,提出一种以句子中心词为基准衡量词语组合相对位置偏移量的句子相似度计算方法。根据词语词性和语法规则确定句子中心词并剔除全局性限定词,以句子中心词为基准对词语相对位置进行标定并计算词语组合的相对位置偏移量,综合句子长度差异性信息、浅层次结构信息和语义信息计算句子相似度。实验结果表明,与基于统计与基于偏移量的句子相似度算法相比,该算法具有较高的相似度计算准确率、召回率及综合指标F值。

On the basis of calculating the semantic similarity of words,a sentence similarity calculation method is proposed to measure the relative position offset of word combination according to the sentence center word. Firstly, central word of the sentence is determined by part of speech and grammatical rules, and the global qualifiers of the sentence are eliminated. Secondly,the relative position of words is marked according to the center word, and the relative position offset of the combination of two words is computed. At last,the sentence length difference information,shallow structure information and semantic information are integrated to calculate sentence similarity. The experimental results show that the algorithm can effectively improve the accury, recall rate, and F value of the comprehensive index compared with the sentence similarity algorithm based on statistics and the one based on offset.