基于最近邻算法的短语结构语法关系判定方法

Judgment Method of Phrase Structure Grammar Relations Based on Nearest Neighbor Algorithm

提出了一种基于词义相似度和最近邻算法的“N1+N2”结构短语语法关系判定方法。首先从“N1+N2”结构中两对名词间的语义相似度定义了短语结构间的相似度,在此基础上给出最近邻分类算法所需要的短语结构间距离的概念;然后建设了一个标注了词语语义类别和短语语法关系的“N1+N2”结构的样本语料库,建立了一种能够标注“N1+N2”结构关系的最近邻分类算法;最后用测试集中计算机标注结果与人工标注结果比较来测试算法效果。实验结果显示,基于论文所提算法的计算机自动标注结果正确率达到97.55%,该结果证明了论文设计算法的有效性。

A method to determine the grammatical relationship of"N1+N2"structure based on word semantic similarity and nearest neighbor algorithm is proposed.Firstly,the semantic similarity between nouns in"N1+N2"structures is defined.On this basis,the concept of distance between phrase structures headed by nearest neighbor classification algorithm is given.Then,a sam⁃ple corpus of"N1+N2"structure is constructed,which annotates the semantic category of words and the grammatical relationship of phrases,and a nearest neighbor classification algorithm that can label the structural relationship of"N1+N2"is established.Fi⁃nally,the algorithm is tested by comparing the results of computer annotation and manual annotation in test set.Experimental re⁃sults show that the accuracy of the proposed algorithm is 97.55%,which fully proves the effectiveness of the proposed algorithm.