Collaborative Filtering (CF) is a frequently used recommendation method that has been successfully applied on many real-world systems. An important problem in CF is selecting a proper set of users and employing them in the rating prediction. Most of current similarity metrics incorporate only the rating values and ignore other information sources in their calculations; while in most of real-world systems, there are sparse data. To tackle this issue, in this paper, a two-step CF method called TCFGA is proposed to provide a set of similar users. In addition, their importance is weighed by employing social trust statements along with rating values. To evaluate the effectiveness of TCFGA, several experiments are performed on two real-world datasets and obtained results show the superiority of TCFGA compared to the others.
