Abstract: Nowadays, the Internet not only becomes an important platform for the public to obtain information and express views, but also brings the risk of social public opinion events. By predicting the trend of network public opinion in advance, the development of hot events can be accurately judged to provide suggestions to relevant departments of government for dealing with public opinion crisis. To solve the problems of poor prediction of single prediction model and great influence of social media on the trend of public opinion, a public opinion prediction method was proposed based on microblog hotspot analysis and LSTM neural network. The network public opinion prediction system was constructed for public opinion time series data analysis by web crawler and PyTorch machine learning platform. Considering the strong currency of microblog, the microblog heat score was calculated with the network hotspot analysis technology.LSTM network was improved, and MH-LSTM prediction model with two hidden layers was designed. Applying MH-LSTM model into the quantitative prediction of Baidu index of public opinion events, the experiments show the correctness with good prediction effect of the proposed model.