In this paper, a stock trend forecasting model is constructed based on Bert’s text sentiment analysis and the forecasting method of LSTM. In order to improve the traditional forecasting model, which does not take into account the influence of market sentiment on stock prices, we use Bert’s model to extract textual information features from social media information, market news, and stockholders’ comments after using historical stock trading data as features in the model for forecasting and carry out text sentiment analysis. The text features are then combined with historical stock data, and the fusedmax function is used to filter out the most likely outcomes to predict stock trends.