After applying the existing information intelligent fusion methods, the redundancy of information data is high. Therefore, in order to reduce information redundancy, this paper designs an intelligent fusion method for social media news information based on reinforcement learning. First, different types of original social media news information data are processed to obtain the basic trust distribution function BPA of the information data, and obtain the data characteristics. Then, the reinforcement learning method is used to capture the relationship between different data features, and the interaction between multimodal features is realized through bilinear pooling. In order to achieve stable multimodal representation of social media news information, information from different models needs to be integrated. Finally, the fusion results are output through neural network training and learning. The experimental results show that the redundancy of social media news information data is less than 32Byte after the application of the method in this paper, which controls the balance between redundancy and data quality, indicating that the method is effective.