A Study on the Prediction of Stock Market Returns Based on Financial News Sentiment Combined with NLP and Deep Learning

Abstract

The complexity and volatility of financial markets have made stock return prediction a research hotspot. Traditional prediction methods often rely on historical price data, neglecting the key nonlinear influencing factor of financial news sentiment. The high proportion of retail investors in my country's stock market and their susceptibility to sentiment further highlight the importance of this factor. This study aims to construct a financial news sentiment analysis framework integrating Natural Language Processing (NLP) and deep learning to accurately predict stock market return volatility trends. The study collected 150,000 financial news articles and corresponding stock trading data from 2018 to 2023. After preprocessing, sentiment features were extracted using a financial sentiment dictionary. Comparative experiments were conducted using logistic regression, random forest, and Long Short-Term Memory (LSTM) networks, and the model was optimized by incorporating temporal features. The results show that the LSTM model has the best prediction performance, with an F1 score of 0.82 and a root mean square error (RMSE) of 0.035, significantly outperforming traditional models. News sentiment and stock returns show a significant nonlinear correlation; negative sentiment has a stronger impact on market volatility than positive sentiment, and different types of stocks exhibit different sensitivities to sentiment. This study provides quantitative support for investors to formulate scientific investment strategies and for regulatory authorities to implement risk management, promoting the in-depth application of sentiment analysis technology in the capital market.