基于知网的新词语相似度算法研究

New Word Similarity Algorithm Research Based on HowNet

基于"知网"提出了一种新的词语相似度计算方法。在概念层次上,引入义原类相似度的概念及计算规则,结合词语概念中主要义原类限制次要义原类和变系数法对各义原类加权计算,求得概念相似度;在词语层次上,引入词性相似度的概念,取不同词性的最大值作为词语相似度。实验结果表明,与已有方法相比,该方法有效提高了词语相似度的精确度和计算效率。

A new word similarity algorithm based on How Net semantic lexicon is proposed in the paper.In the conceptual level, this paper introduces the concept and calculation rules of sememe class similarityto do weighted calculation on each sememe class, combining the idea of the main sememe class limitingthe secondary sememe class and variable coefficient method,with this method, the conception similaritycan be achieved; in the word level, the concept of the similarity of the parts of speech is introduced, andthe maximum value is taken as word similarity. The experiment results show that the new method effective-ly improves the accuracy and computational efficiency of word similarity, compared with the existing meth-ods.