Stock Market News Sentiment Analysis and Trend Prediction Using Transformer Models

Abstract

This study aims to use the Transformer model to achieve sentiment analysis and trend prediction of stock market news. As the financial market becomes increasingly sensitive to changes in news sentiment, traditional prediction methods based on historical price data have difficulty accurately capturing the complexity of market fluctuations. To this end, we apply the Transformer model to news text sentiment analysis, and capture the emotional features in news texts through a multi-head self-attention mechanism, thereby effectively identifying the changing trend of market sentiment. In the experiment, we combined the sentiment score with the stock price time series data to construct a sentiment-driven prediction model to predict future market trends. The results show that the Transformer model has high accuracy and robustness in sentiment analysis and trend prediction tasks, significantly outperforming traditional machine learning models. This study provides new ideas for the application of sentiment analysis in financial forecasting and proves the potential of multimodal methods combining sentiment information and price data in financial market analysis.