Word Sense Disambiguation(WSD) plays an important role in many areas of natural language processing. In order to deal with large scale WSD, an unsupervised WSD method is provided in this paper. Based on the vector space model, the mapping relationship between word sense and sense categories is set up by using Machine Readable Dictionary (MRD) and sense categories thesaurus; and then use the knowledge of sense categories to learn disambiguate feature in corpus unsupervi ̄sedly; finally, the di...