Recommendation systems provide a solution to tackle information overload problem. These systems have several limitations, one of which is cold-start users. In this article, a new method is proposed to overcome the cold-start user problem. The main idea of this study is to apply a clustering technique using trust relations and rating information to compute the weights. First, the implicit relations are determined, and then the similarity is computed for each pair of explicit and implicit relations. Second, confidence values are determined through an information rating by dividing the number of common items for each pair of users by the number of items that have been rated by the first user of this pair. Furthermore, the similarity and confidence values are integrated to produce weight values, and then the distance values are inferred. Additionally, the partitioning around medoids clustering algorithm is adopted to cluster the users into groups according to their computed distances. Moreover, the Singular Value Decomposition Plus (SVD++) method is employed for each cluster to predict the items for cold-start users. Eventually, the proposed method is evaluated with two real-world datasets. The results reveal that the proposed method outperforms the state-of-the-art trust methods in terms of prediction accuracy.