An Algorithm For Optimizing Word Similarity In "Knowledge Network"

Natural language processing (NLP) is an important and vibrant field of interdisciplinary Computer Science research. Natural language words have a complex relationship. We need a way to measure,in other words,need to find a way to put one complex relationship with a simple number to measure in practical application.And,the word similarity is one of them.In many fields,such as word sense disambiguation,information retrieval,example-based machine translation and so on,Word similarity is broadly used.Now,mang of word similarity algorithms are based on vector space model(VSM).But, we will find these ways cause lots of problems of high dimension and sparseness in practical application.However,these ways can not sovle two important problems:synonym and polyseme. The traditional word similarity calculation only considers the distance between the original words, but also only consider the relationship between the upper and lower bit words,,without considering other relationships between the original meaning.What’more,similarity calculation does not consider a different set of a pair of words which is different between in word similarity, and the concept of similarity calculation does not take full account of the relationship between the four original meaning,thus make the computation complicate.I will try to think differently in this paper and discuss the existing algorithms deeply.What’s more, I’ll give a new algorithm based on How-net and develop a software to calculate word similarity.The test results show that my algorithm have more excerlent accuracy than traditional algorithm.