An Improved Chinese Word Semantic Similarity Algorithm Based on CiLin

摘要

The CiLin is a famous semantic dictionary of Chinese synonyms; its structure and function are quite like the WordNet in English. This paper improves the existing algorithm of Chinese word semantic similarity based on CiLin, which integrates the word distance, the density of lowest common parent node and branch layer spacing. Firstly, the initial value of word semantic similarity is calculated through word distance, and then an adjusting parameter that depends on the lowest common parent node density n and the branch interval k is set to revise the initial value downward. Through the fourth root of an expression for the parameters k and n, the revision range of initial similarity can be limited below 16%, thus avoiding the unreasonable phenomenon that the word pairs with near distance have a low similarity because of a far branch interval. This method obtains an as high as 0.8464 value of Pearson correlation coefficient compared with artificial judgment for the word pair set of Miller & Charles.