Collaborative filtering is the basis of the recommendation system. Collaborative filtering generally finds users (neighbors) with the same interests as the target users among a large number of users. How to determine whether the users have the same interests as the target users and how to form the neighbors of the target users. Sorting directories has become a major issue. However, there are some shortcomings in the existing neighbor selection scheme. For example, when calculating the deviation of two users from the same group of items ratings, the existing two-layer neighbor selection scheme only considers the sum of the individual items rating differences. It is unfair to users who are not much different from the target users. We propose a trustworthiness calculation scheme based on Kendall correlation coefficient and standard deviation. Specifically, when the sum of the difference between the two neighbors and the target user is the same, the trustworthiness of the user with high Kendall correlation coefficient and small standard deviation is higher.