A collaborative filtering algorithm based on user trust ranks and items is proposed to improve the anti-“shilling attacks” ability. Firstly, a user relationship graph is built based on user interest similarities, rating similarities, and rating correlations. Secondly, using the relationship graph, a userrank model is proposed to calculate user trust ranks. Thirdly, the userrank values are taken as users’ weights to incorporated into the typical item-based Slope One algorithm. Finally, we experimentally evaluate our approach and compare it to Slope One. The experiment results suggest that our approach provides better recommendation than Slope One.