Authors: Harsha Sammangi, Aditya Jagatha, Hari Gopal Maddireddy
Abstract: This study presents a sentiment-driven Decision Support System (DSS) that leverages advanced word embedding techniques—Word2Vec, GloVe, and BERT—to analyze CEO earnings call transcripts and predict stock market reactions. Tra- ditional lexicon-based sentiment models fail to capture the nuanced, contextual language used by executives. By employing pre-trained embeddings and machine learning classifiers, the study enhances the accuracy of sentiment classification. The proposed system integrates quantitative sentiment scores with event study method- ology to assess the impact of CEO tone on stock performance. Thematic analysis further enriches interpretability by identifying recurring patterns in executive com- munication. Results demonstrate that positive CEO sentiment generally correlates with stock appreciation, while negative sentiment aligns with declines. Among models tested, BERT outperformed others in classification accuracy. This research contributes to real-time financial analytics by embedding sentiment intelligence into DSS frameworks, supporting investors, analysts, and automated trading sys- tems with improved decision-making capabilities grounded in contextual linguistic analysis.
DOI: https://doi.org/10.5281/zenodo.15845150
