Joint Topic-Emotion Modeling in Financial Texts: A Novel Approach to Investor Sentiment and Market Trends

This paper presents a novel approach to financial text analysis by jointly modeling topics and emotions within financial news and social media discussions, thereby advancing market trend understanding and investor sentiment analysis. Leveraging the Neural Variational Document Model (NVDM) for topic modeling, we effectively capture the thematic structure across global financial texts, achieving a coherence score of 0.68 and a perplexity of −8.5. For emotion classification, a fine-tuned FinBERT model yields an accuracy of 96.3%, F1 Score of 97.8%, precision of 98.3%, and recall of 97.5%. Our integrated framework not only uncovers the latent topics driving market discourse but also associates these topics with nuanced emotional states, providing a comprehensive perspective on financial sentiment. This joint modeling approach enhances the interpretability and reliability of financial market predictions, offering a scalable solution for understanding investor emotions with significant potential for applications in market analysis, trading strategies, and economic forecasting.