Sentiment Analysis of Financial Texts for Predicting Digital Market Trends Using Transformer-Based Deep Learning Models

This study investigates the effectiveness of Transformer-based deep learning models in financial sentiment analysis with the aim of improving the accuracy of digital market trend prediction. The research utilizes a dataset of 5,842 financial text entries categorized into positive, neutral, and negative sentiments derived from various financial news, corporate reports, and market analyses. The data were preprocessed and analyzed using both traditional machine learning algorithms, including Logistic Regression, Support Vector Machine (SVM), and Random Forest, as well as a fine-tuned Bidirectional Encoder Representations from Transformers (BERT) model. The baseline models achieved moderate performance, with Logistic Regression producing the highest accuracy of 71.6 percent, while the fine-tuned BERT model outperformed all traditional approaches, achieving an accuracy of 78.8 percent and higher F1-scores across sentiment classes. These results indicate that BERT’s contextual embeddings effectively capture semantic nuances and subtle tone variations in financial language, which are often overlooked by conventional models. The confusion matrix analysis revealed that most misclassifications occurred between neutral and negative classes due to the cautious tone of financial writing, yet BERT remained robust in identifying overall sentiment direction. The study concludes that Transformer-based models provide a more reliable and context-aware framework for analyzing financial texts and have significant potential for applications in digital market sentiment prediction, AI-driven financial analytics, and automated decision-support systems in trading environments.