基于节点词全句共现的动态词义消歧研究

A Study of Dynamic Word Sense Disambiguation Based on Full-sentence Co-occurrence of Node Word

文章根据词义消歧即将词义回归语境这一特性,提出了一种基于节点词全句共现的动态词义消歧方法。该方法首先以全句为窗口限定节点词的使用语境,其次使用互信息(MI)、卡方检验(χ^(2)检验)和相对词序比(RRWR)等统计方法抽取节点词的语义相关词,并参照《同义词词林》构建相关词语义范畴库,最后以共现频数作为加权系数,依靠单义词语义聚类分布率对中低频共现多义词进行消歧。采用该方法对与“美丽”共现的1030个小于7义类的多义词进行消歧的测试试验中取得了85.2%的正确率。

Based on the property that word sense disambiguation is the return of word sense to context,we propose a dynamic word sense disambiguation method based on full-sentence co-occurrence of node word.The method firstly uses the full sentence as a window to limit the node word usage context,secondly uses statistical methods such as mutual information,chi-square test and ratio of relative word rank to extract semantically related words,and builds a related semantic category database by referring to“Tongyici Cilin”(A Dictionary of Synonyms),and finally uses the co-occurrence frequency as a weighting factor to disambiguate the low and medium frequency co-occurring multisense words by relying on the distribution rate of single-sense word meaning clusters.The method is used to disambiguate 1030multiple-meaning words with less than 7meaning categories that co-occurred with“meili”(beautiful),and a correct rate of 85.2%is achieved in the test.