<![CDATA[In the age of digital transformation, the development of digital banking platforms such as BRImo by Bank Rakyat Indonesia (BRI) continues to evolve to improve customer experience. However, many users still express dissatisfaction through online reviews, especially on platforms such as the Play Store and Twitter (X). This study conducts a systematic and fair comparison between a traditional machine learning approach (Naïve Bayes) and a deep learning approach (Long Short-Term Memory) for sentiment classification under identical dataset conditions. User reviews were collected using web scraping and crawling techniques, followed by text preprocessing, lexicon-based labeling, and appropriate feature representations for each model. The results indicate that both algorithms classify sentiments into three categories: positive, negative, and neutral. The Naïve Bayes model achieved an accuracy of 89%, with macro-average precision, recall, and F1-score of 0.88, 0.58, and 0.59, respectively. Meanwhile, the LSTM model achieved an accuracy of 85%, with macro-average precision, recall, and F1-score of 0.59, 0.63, and 0.60. The findings reveal that Naïve Bayes demonstrates more stable performance on short and highly imbalanced user review data, while LSTM shows limited improvement for minority classes despite its contextual modeling capability. These results highlight the importance of dataset characteristics and evaluation metrics beyond accuracy in sentiment analysis tasks. This research provides practical insights for BRImo development teams and contributes to the understanding of model behavior under real-world sentiment data imbalance. ]]>
Copyrights © 2026
