With the development of internet technology, the personalized tag recommendation system plays an important role in information or resources filtering. In Sina microblog website, an user can freely tag himself to indicate his interests. Meanwhile, users can also search other users who have the similar interests through tags. For the issue that there are no tags or few tags for the most users in Sina microblog website, an algorithm based on RBLDA model and users' interaction graph for tags recommendation is proposed in this paper. The algorithm utilizes the RBLDA model to produce the intial list of tags, and combines with users' interaction graph generated from actions of interaction between users to predict the final tags. The experimental results carried on some real data sets show that the proposed method performs better than traditional tag recommendation algorithms in comparison.