The Application of Deep Learning and Artificial Intelligence for Media Influence of News Portal

This article presents a model that provides technical support for managing public opinion and coordinating communication strategies on news platforms. Based on the robustly optimized, bidirectional encoder representations from transformers pretraining approach pre-trained language model, this study extracted deep semantic features from news texts to enhance the ability to understand and classify news content. By integrating user behavior data, a heterogeneous communication graph of “content–platform–user” was constructed, using a graph attention network to achieve structured modeling of multi-hop communication paths. Finally, a multi-task joint optimization framework was designed to fuse content features, user behavior, and communication path information—forming a multi-dimensional influence evaluation index system. This system included communication coverage rate, path restoration rate, user response consistency, and emotional impact index.