The definition of semantic relevancy is ambiguous and non-uniform to the chinese semantic relevancy computing model for the present. and semantic relevancy computing in always measured by semantic similarity, which has led to limitations in the application of semantic relevancy computing. choosing the relationship between words as the study object, using mutual information (MI) to present the interrelationship between words, it is proposed a new semantic relevancy computing method based on corpus statistics. the vector space is used to store the data and the result, and the algorithm effectively improves the time and space complexity. the result conforms to the cognitive styles of reality, which verifies the validity of the method effectively.