In the traditional collaborative filtering recommendation algorithm, similarity calculation is the core of the algorithm. However, the previous calculation method is too dependent on the user’s score, does not consider the user’s own attributes and trust relationship, and does not distinguish malicious users. In order to solve the appeal problem, this study introduces an improved new trust relationship measurement method into similarity calculation. This new method not only considers the influence of malicious users, but also combines the properties of users effectively. In addition, the study also improves the similarity algorithm on the hot issues. The algorithm finally uses the initial user clustering to get the adjacent users, effectively eliminating the cold start and data sparsity. In the experimental part, it can be proved that the proposed algorithm can effectively improve the recommendation accuracy by comparing with other algorithms.