Abstract: The current computation models of Chinese semantic relevancy variate due to various definitions of relevancy degree.These models are almost based on a similarity computation model，which much limits its application.This paper presents a new definition of semantic relevancy，and defines two words semantically related if the two concepts associated with the two words respectively have the relationship in a semantic knowledge base like HowNet.By mining all the direct and indirect relationships，a new computation model of semantic relevancy is proposed.This model can be generalized to any semantic knowledge bases constructed by the knowledge dictionary mark-up language.Finally，it applies the computation model to the word sense disambiguation problem，and therefore validates that this computation model can be efficiently employed to many applications.