Abstract: Determining the frequency of drug side effects is a key issue in drug development and risk-benefit evaluation, which is usually achieved through randomized controlled clinical trials, but this method has limited performance and is expensive. As data from drug clinical trials increase, it is possible to study side effects based on data-driven computational methods. The existing drug-side effect association prediction methods mainly explore the drug-side effect association. In contrast, this paper aims to study the prediction of the frequency of drug-side effects, and quantitatively analyze and compare each side effect of each drug according to the frequency of side effects in clinical trials of drug development. In this paper, a cooperative Propagation model FPDSF (Rating Propagation Model for Predicting Frequencies of Drug-Side) based on drug-side high order similarity network is proposed. The model builds biological entity similarity networks based on known drug-side effect frequency information, and deduce the frequency of potential drug side effects through the process of high-order collaborative propagation of known frequency information in the network. In addition, a similar network improvement method based on neighborhood learning is proposed to improve the prediction performance of the model. Experiments are carried out on the real drug-side effect frequency data set obtained in SIDER 4.1 and ADRecS 3.1. Compared with the existing optimal method, the proposed method decreases by 6.98% and 7.23% in the mean square error and mean absolute error indicators respectively.