As the core of natural language processing, text classification has become a key research issue in text processing. This paper focuses on the prediction of large-scale and common diseases in society. Obtain news texts reported by major news media around the world, respectively count the top ten disease rankings in the news texts, and analyze the characteristics of the original data distribution. In this paper, the text model based on CNN and LSTM networks and the disease trend model based on LSTM networks are merged to comprehensively analyze the text information of news and the time series of diseases, and use a special strategy for selecting diseases. The experimental results show that the strategy achieves more than 70% accuracy on seven different news data sets. The fusion strategy and disease selection strategy proposed in this paper have certain significance for disease trend prediction and can help improve the accuracy of disease trend prediction.