Financial Sentiment Analysis

The field studies how news, social media, corporate disclosures, and central-bank communication express sentiment and other market-relevant signals. Common approaches include financial lexicons, supervised models such as FinBERT, fine-tuned or prompted large language models, and methods that capture aspects, uncertainty, emotion, or the sequence of language. Researchers evaluate these systems on labeled text and downstream tasks such as return prediction, risk detection, and trading, sometimes combining text with prices or other market data. A central tension is that strong sentiment classification does not reliably produce robust profits: results can vary by asset, language, time horizon, and market regime, and credible evaluation must account for timing, transaction costs, and changing conditions.