A Comprehensive Review and Empirical Results on News Sentiment–Driven Machine Learning Models for Stock Market Price Prediction

Abstract

This paper presents a comprehensive review and empirical evaluation of news sentiment–driven machine learning models for stock market price prediction. The study synthesizes existing research on financial sentiment analysis, highlighting the growing relevance of textual signals derived from news headlines and articles in forecasting short-term market movements. Leveraging multiple sentiment extraction techniques—including lexicon-based approaches, machine learning classifiers, and transformer-based models such as FinBERT—the research integrates sentiment with historical price data to assess predictive accuracy across various model classes. Empirical experiments compare baseline statistical models, traditional ML algorithms, and deep learning architectures, revealing that sentiment-augmented models consistently outperform price-only approaches, particularly for short-horizon predictions. Results from robustness checks and backtesting further demonstrate the practical value of sentiment-enhanced forecasting in real trading environments. The study contributes a unified framework, benchmark results, and insights that advance sentiment-aware financial prediction research.