Prediction of stock market volatility based on public opinion mining and multi-source data fusion: A case study of Guoxuan Hi-Tech

Abstract

This study aims to enhance the accuracy and robustness of stock market volatility forecasting in the context of rapidly evolving public opinion and heterogeneous data environments. Traditional models often struggle to capture the nonlinear and dynamic interactions between investor sentiment and market behavior, particularly during high-volatility periods. To address these limitations, a unified deep learning framework is developed that incorporates multi-source sentiment signals into the forecasting process. Specifically, a Transformer–LSTM threshold fusion model is constructed to extract semantic features from investor comments and temporal patterns from historical market data. The model introduces a dynamic fusion mechanism based on historical volatility thresholds and integrates a dual-dimensional sentiment indicator system (CAI and CSEI) to reflect behavioral signals from both current and potential investors. Empirical evaluation using Guoxuan Hi-Tech stock data from 2019 to 2024 shows that the proposed model significantly outperforms benchmark methods such as BERT–GRU and LDA–Prophet in predictive accuracy, generalization, and robustness. These findings underscore the effectiveness of integrating deep learning with multi-source sentiment modeling and suggest a scalable framework for broader applications in financial risk forecasting.