In recent years, social tags have become useful tools for improving the accuracy of recommendation and be a basis of personal recommendation. In this paper, we study the concept of users’ behaviors on objects to define top-k similar objects pairs from a couple of users’ browsed object sets recently, the objects pairs will represent the users’ behavior similarity recently. We develop a Set Join method to find similar object pairs, and use Index to speed up the process. From the experiment we find latter time is spent one-fifth of the former.