一种改进的基于《知网》的词语相似度计算方法

Modified word similarity computation approach based on HowNet

《知网》是一部比较详尽的中文语义知识词典,共用1618个义原描述词语,故相关的词语用《知网》的概念描述时,有相同的义原。通过这一规律,与当前的词语相似度计算方法结合,提出改进的方法计算相关词对的相似度。并引入弱义原的概念,排除弱义原对词语相似度计算的干扰。实验证明:该改进方法更符合人的直观,更适用于文本挖掘。

HowNet is a lexical base with rich semantic information. It uses 1618 sememes to describe words. The related words have the same sememe when they are described by the HowNet. Combined with the current computation algorithm of the words＇ similarity, the paper proposed an improved algorithm to compute the similarity between the related words. It also introduced concept about weak sememes and excluded such sememes＇ interference when they appeared in the computation of the word＇s similarity. The experiment proves the improved word similarity computation meets the peoples＇ intuition and text mining better.