An important facet of Last.FM and MovieLens is that users manually annotate the items using so called tags. There are many researches about using tag to improve the quality of recommendation. However, tags are “local” descriptions of items given by the users,because different people use different tags for the same item, but the tags may represent the same means. In this paper, we used tag grouping method to group the tag according to the similarity of co-occurrence distributions. Based on this, we proposed an approach to group synonymy tags and fusing the relationship between users-tag with the collaboration filtering algorithms. The results of the empirical evaluation show that the approach is effectiveness in augmenting recommendation.