This paper analyzes the network expression and audience participatory discourse system under the mainstream media communication mode, combines the audiovisual multimodal audience cognitive participation and the multimodal features of the mainstream media, and proposes a multimodal deep learning model based on mainstream media communication. Based on the picture vision and human emotion migration methods, the picture vision feature migration module and the text emotion migration module are constructed respectively, and the multimodal feature fusion model is built by applying the gradient enhancement decision tree. Linguistic modality analysis of mainstream media communication discourse is carried out by analyzing the development characteristics of audience participatory cultural communication modes. The influence of different mainstream media communication discourses on the audience’s participatory culture is examined through logistic regression analysis while screening network discourse access. The regression analysis shows that the mainstream media communication modes represented by social software and microblogging have a significant impact on the development of the audience’s participatory culture, with p-values of 0.0452 and 0.0068, respectively, and thus the mainstream media need to strengthen content innovation, focus on specific audience groups, and make full use of new media platforms to provide audiences with richer, more accurate and professional news services.