Sentiment Analysis is a process where a text is assigned whether it has a positive tone, a negative tone, or a neutral tone. In finance, sentiment analysis is used to derive a numerical signal out of financial text. In order to do financial textual analysis on a collection of financial texts that are a component of a larger dataset called the Financial Phrase Bank, this study presents a novel strong sentiment classification architecture based on an improved model of Bidirectional Encoder Representations from Transformers (BERT). The methodology involves all preprocessing requirements, which are text cleaning, BERT Word Piece tokenization, label encoding, and down-sampling to balance the classes. Common measures, such as accuracy of 95.29%, precision of 95.37%, recall of 95.24%, and f1-score of 95.32%, were used to train and evaluate the model. These metrics show that the model performs better than more traditional machine learning models like Random Forest. together with language models like GPT-4o. Visualization was used to explain the model behavior and generalization power; e.g., the visualization of a confounding array, training-validation curve plots, and a token distribution plot. The experimental findings justify the utility of the proposed BERT-based approach in retrieving subtle sentiment expression in elaborate financial narrative texts, thus providing scalable and effective accuracy to sentiment-based financial analytics
Sandeep Gupta (Fri,) studied this question.