In lots of applications, the traditional item-based collaborative filtering algorithm usually has excellent performance on prediction accuracy and recommendation instantaneity, but it has poor recommendation precision. In order to solve this problem, this paper, through the experimental results, analyzes the causes of the problem, and points out the shortcomings of the candidate item set selected by traditional itembased collaborative filtering algorithm. And then proposes a new method to select candidate item set based on relations between items. On that basis, we introduce the method into tr...