Research on Time Series Forecasting Based on Text Information and BP Neural Network

With the continuous development of the global economy, capital markets have also been evolving and improving. In recent years, the advancement and innovation of information technology have gradually drawn investors’ attention to quantitative investment. Additionally, with the widespread adoption of mobile internet technology, textual information such as financial news and stock forum comments has become a crucial basis for most investors’ decision-making, influencing stock price trends to some extent. Against this backdrop, in-depth research on quantitative investment strategies from the perspective of textual information holds significant implications for both the market and investors. This paper employs text mining methods to design a stock sentiment index. Initially, a BP neural network without incorporating the investor sentiment index is constructed to predict the closing price. Subsequently, the designed stock sentiment index is integrated into the BP neural network to build a prediction model for forecasting the closing price. Finally, the predictive performance of the two models is compared. The research results demonstrate that, by comparing the mean squared error (MSE) of the BP neural network predictions, the neural network incorporating investor text sentiment achieves higher prediction accuracy for stock closing prices compared to the one without investor text sentiment.