FINANCIAL SENTIMENT ANALYSIS USING FINBERT FINE-TUNED

Abstract

Financial information is a critical type of data for analysis. However, because much of it is unstructured and widely dispersed, an appropriate analytical method is required, one of which is sentiment analysis. In the financial context, sentiment analysis is employed by the industry to assess public perceptions of companies or market conditions. This study implements a fine-tuned FinBERT model to perform sentiment analysis in the financial sector. The dataset used is a combination of FiQA (Financial Question Answering) and The Financial PhraseBank, consisting of English sentences labeled with negative, neutral, and positive sentiments. The research process involved data preprocessing, tokenization, data splitting, model training, and evaluation using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results show that the model achieved 82% accuracy, with its best performance in the positive class (F1-score 0.88) and the neutral class (F1-score 0.85), but weaker performance in detecting the negative class (F1-score 0.49). These findings indicate that the fine-tuned FinBERT is effective for financial sentiment analysis, particularly for positive and neutral sentiments, though improvements are needed in negative sentiment detection, potentially through expanding training data diversity or applying data augmentation techniques