In this paper, we design a method to define the degree of tag ambiguity, and clarify ambiguous tag by clustering it into different groups, each group contains an explicit concept. Furthermore, based on the vote strategy, we design an algorithm to detect the group that a tag belonging to in order to solve tag ambiguity. Besides, according to this concept, we provide a novel framework of tag recommendation system, insteading of finding relation between tag and tag, our new framework find relation between tag and tag cluster. By this method, we take issue of tag ambiguity into account and improve the recommendation result.