摘要： WordNet在计算概念相关度时存在词汇量小、难以及时扩展更新以及同义、近义、一词多义等问题。为此，提出一种结合文本信息和图像视觉信息的概念相关度方法。利用Google距离和KL距离分别计算基于词语同现频率的概念相关度和基于视觉特征的概念相关度，并结合两者得到概念的总体相关度。实验结果验证了该方法的有效性。

关键词: 概念相关度, WordNet网络, Google距离, KL距离, 视觉语言建模

Abstract: For WordNet has its limitation of the vocabulary and promptly expansion, the problem of the synonym, near-synonym and polysemy, this paper proposes a new concept similarity measurement algorithm, which combines text and image vision information. Google distance is used to calculate the semantic distance between the concepts based on co-occurrence of the words. And the visual similarity is calculated. The whole similarity between the concepts is measured by combining them. Experimental results verify the validity of this algorithm.

Key words: concepts similarity, WordNet, Google distance, Kullback-Leibler(KL) distance, vision language modeling

中图分类号: 

• TP301.6