Sense decomposition from E-HowNet for word similarity measurement

In this paper, a novel Chinese word similarity measuring algorithm is proposed. It utilizes the information in Chinese dictionaries and decomposition of all the word definition in E-HowNet into semantic attributes which include hidden relationships and meanings. The extracted semantic attributes are regarded as semantic vectors for word similarity measurement. Evaluation results not only show that our word similarity measuring algorithm can achieve a higher correlation with human annotation than other methods, but also can successfully solve the unique edge distance and indistinguishability problem in taxonomy-based word similarity measurement.