In the era of big data, the news form of cultural communication is closely related to its dissemination depth and coverage. This paper takes news information as the research object and explores its morphological innovation path based on neural network algorithm. The characterization method of complex network is briefly analyzed as the theoretical basis for the research. Two commonly used key node description indexes, namely, compact centrality and median centrality, are successively described, and the PageBank algorithm is adopted to identify the important nodes in the process of network news dissemination. Considering the multimodal information in the process of news dissemination, a variant of BERT model (HFBM) based on the Transormer model is introduced, and a convolutional neural network is used to extract the modal eigenvectors of tagged words of news information. The model performs classification and prediction of news information sentiment by fusing modalities and their corresponding feature vectors. Combining the above, the model method design of news feature mining and prediction based on BERT model is completed. At the same time, it elaborates the characteristics of news dissemination in the era of fusion media, and combines the proposed news feature mining and extraction algorithm to realize the innovation of the news form path of cultural dissemination. In the news forwarding volume prediction module constructed based on the proposed algorithm, the algorithmic model of this paper always maintains better prediction results during the whole information dissemination process (within 15h). It shows that the designed news feature mining and extraction algorithm is able to accurately model and predict news dissemination trends.
