关键词: 词语相似度, 知网, 义原, 义项, 词性

Abstract: The word similarity computation is widely used in text classification, question-answer system, machine-translation and text clustering. Research of this computation is generally based on HowNet, according to the distance and the depth of sememes. Based on above, an improved method of word similarity computation was proposed as follows. Firstly, a new variable coefficient of homonym similarity computing was proposed according to the count of homonym. Secondly, it took part of speech into account and argued that the part of speech of homonym is different in contributions to word similarity and removed the combinations of homonyms with different part of speech. The experimental results show that the result obtained through this newly-improved computation method is better with less complex calculation and higher calculation efficiency.

Key words: word similarity, HowNet, sememe, homonym, part of speech