According to the problem of the low precision rate of resources recommended in O2O E-Commerce platform, the paper calculates user's global-trust in the system from two aspects of user activity and user authority. On this basis, it improves the traditional collaborative filtering algorithm with the introduction of trust relationship between users. It sets confidence threshold to determine the scope of neighboring users. On this basis, combined with trust and similarity to determine the neighbor users, this work puts their combined mixed Numerical as recommended weights. The experimental results can prove the validity and superiority of the proposed algorithm.