Measuring Investor Sentiment in Financial Discourse: How Different Approaches Shape Market Signals

Abstract

Stock prices are shaped not only by fundamental data but also by investor sentiment, which often deviates from rational decision-making. Given the vast volume of financial texts published by both professional and amateur investors—especially on online financial platforms—sentiment analysis in such unstructured data is essential to understanding their impact on market movements. This study examines the interplay between text data and stock market movements, highlighting the critical role of sentiment extracted from financial news and online news. Existing research has largely relied on general-purpose lexicons or uniform classification techniques, which limits the accuracy of sentiment analysis in financial contexts. To address this gap, we propose a hybrid framework that integrates domain-specific lexicons with advanced machine learning classifiers to improve sentiment extraction from unstructured financial text. Our approach evaluates the impact of lexicon selection on sentiment scores and examines the relationship between classifier choice and prediction accuracy. By improving sentiment analysis methodologies, our findings contribute to the development of more robust stock market forecasting models, strengthen decision-making processes for investors, and increase market efficiency.

Stock prices are shaped not only by fundamental data but also by investor sentiment, which often deviates from rational decision-making. Given the vast volume of financial texts published by both professional and amateur investors—especially on online financial platforms—sentiment analysis in such unstructured data is essential to understanding their impact on market movements. This study examines the interplay between text data and stock market movements, highlighting the critical role of sentiment extracted from financial news and online news. Existing research has largely relied on general-purpose lexicons or uniform classification techniques, which limits the accuracy of sentiment analysis in financial contexts. To address this gap, we propose a hybrid framework that integrates domain-specific lexicons with advanced machine learning classifiers to improve sentiment extraction from unstructured financial text. Our approach evaluates the impact of lexicon selection on sentiment scores and examines the relationship between classifier choice and prediction accuracy. By improving sentiment analysis methodologies, our findings contribute to the development of more robust stock market forecasting models, strengthen decision-making processes for investors, and increase market efficiency.