The growing digital exposure of financial institutions highlights the need for effective reputational risk monitoring. This study proposes the development of a sentiment analysis model applied to texts collected from the X social network (formerly Twitter), aiming to automatically classify user comments based on emotional polarity (positive, neutral, or negative). The methodology includes data collection via API, text preprocessing, class balancing, and the implementation of the Naive Bayes algorithm through supervised learning techniques in Python. Results showed an overall accuracy of up to 69.27% in the original dataset, with performance improvements for minority classes using upsampling. The model proved most effective in detecting negative sentiments, which are crucial in managing reputational risk. The proposed solution is intended to support decision-making in the banking sector by enhancing institutional image monitoring and crisis prevention strategies
Ferreira et al. (Mon,) studied this question.