Aimed at the problems that under some specified conditions, obtaining sufficient samples is so difficult that the accuracy of the BN parameters learned by the maximum likelihood estimation algorithm is often low, and multiple parent nodes collaboration to influence constraints is involved in some areas of practical application, a BN parameter learning method based on the modified multiplicative co-constraint under small data sets is proposed by drawing on the idea of PAVA order-preserving regression algorithm. First, the paper is to determine whether the parameters in the multi-parent part of the known sample data meet the needs of the multiplicative collaborative constraint. Secondly, both the left and right sides not to meet the needs of the multiplicative co-constraint are divided into wholes, and adjusted separately by using the PAVA algorithm. And then, for the adjusted whole, three correction methods with different weights are given to correct each parameter according to the amount of sample data corresponding to the combined state of different parent nodes, and gain mean final parameter learning result. Finally, the proposed method is validated by simulation using a classical grassland wetting network model. The experimental results show that the proposed method not only meets the needs of the multiplicative cooperation constraint under small data set conditions, but also the KL scatter is always lower than the other 2 methods in addition to that the running time is slightly higher than that of the other 2 methods by about 1×10-3s with minimal impact. Generally speaking, the proposed algorithm is superior to the other 2 methods in the comprehensive performance.