Dynamic Prediction Model of News Communication and Advertising Planning Based on AI Gradient Descent Intelligent Algorithm

Predicting the trend of news descriptions through pre-trained language models and neural networks based on intelligent algorithms has always been a hot issue in intelligence analysis. In this paper, we establish a news database that predicts more than 9000 annotated news time trends, filling the gap in the database. An improved method is designed based on the pre-trained intelligent algorithm. In the graph pooling algorithm, combining the Graph U-Nets pooling method with self-attention can better solve predicting news events’ development trends. Experimental results show that compared with the baseline graph classification algorithm, this method has improved results and also solves the problem that pre-trained language models cannot handle super-long text.