Algorithm Design for Stock Price Prediction System Based on Distilbert Financial News Sentiment Analysis Model

As the frequency of information updates in the financial market continues to rise, the traditional stock forecasting methods that rely only on historical prices and macro indicators have been difficult to fully describe the nonlinear impact of news shocks and market sentiment fluctuations. Aiming at the problems of high cost of financial text processing, lack of multimodal fusion, and limited real‐time deployment of prediction system, this paper proposes an end‐to‐end multimodal framework for stock price prediction. The framework uses the lightweight distilbert as the emotional encoder of financial news, reduces the complexity of the model while retaining the semantic expression ability through knowledge distillation, and constructs a dynamic gating fusion module combined with the characteristics of the numerical market, so as to realize the collaborative representation of the text emotional information and the minute level price index. In the prediction layer, spatiotemporal attention mechanism and gating cycle unit are further introduced to model the time series of fused multi‐source features, so as to form a complete system integrating emotion analysis, multimodal interaction, and trend prediction. The experimental results show that the accuracy of the distilbert emotional analysis module on the financial text test set is 88.3, the F1 score is 88.1, and the reasoning speed reaches 3500 texts per second, showing a good balance between accuracy and efficiency. In the system level prediction task, the direction prediction accuracy of this model is 68.7, the annualized rate of return is 24.3, the Sharpe ratio is 2.15, the maximum fallback is 12.8, the single reasoning delay is 85 MS, and the throughput is 235 requests per second. The results show that the proposed method can not only effectively improve the performance of financial text emotion recognition and stock price trend prediction, but also has strong real‐time deployment ability and engineering application value.