An epidemic is a typical public health emergency that refers to the occurrence and rapid spread of disease. A good epidemic transmission model plays a crucial role in preventing an epidemic. The epidemic transmission model is largely similar to the model of sentiment analysis and transmission on social media. Therefore, this paper intend to use the method of deep learning to explore the key issues of theme and sentiment analysis from the perspective of public opinion analysis. In order to fully extract the features automatically, we combine the following methods: multi-channel inputs, multi-granularity convolution kernels, direct connection with high-speed channels, and this paper proposes the multi-channel and multi-kernel (MCMK) model. Furthermore, we leverage generative adversarial nets to combine several single tasks, called Joint-MCMK model, which achieves information sharing and improves the training speed and model accuracy. To verify the validity of our proposed models, this paper experimented with the short text topic classification dataset TREC [1] and the sentiment analysis dataset IMDB [2]. Results achieved 98.6% and 92.6% respectively, which are superior to the highest existing industry benchmark (96.1% and 92.58%). In addition, this paper compared the training spread differences between the joint-MCMK model and MCMK model, which shows that the joint-MCMK model has better performance at training speed. Finally, the control variable method was used to analyze the multiple effects of the different factors. The optimal value of some relevant parameters in our models were verified by several experiments. (C) 2019 Published by Elsevier Ltd.
