Sentiment Classification of Livin’ by Mandiri Reviews in Indonesia Using LSTM for Digital Banking Service Improvement

The rapid expansion of digital banking services in Indonesia has increased the need for continuous monitoring of user satisfaction, particularly through feedback submitted via app reviews. This study analyses user sentiment toward the Livin’ by Mandiri mobile banking application using a deep learning approach. A total of 5,000 user reviews were collected exclusively from the Google Play Store and pre-processed through text cleaning steps such as slang normalization, stemming, and tokenization. Sentiment labels (positive, neutral, negative) were assigned using an Indonesian lexicon-based method, and a Long Short-Term Memory (LSTM) model was trained and evaluated with accuracy, precision, recall, and F1-score metrics. Results indicate negative sentiment dominates (37.6%), with frequent complaints about login failures and slow performance, while the LSTM model achieved 98% accuracy. This study is limited by its single-platform data source, potential linguistic bias in Indonesian user reviews, and the model’s limitations in detecting sarcasm or complex emotions. Nonetheless, the findings demonstrate the applicability of sentiment analysis as a real-time monitoring tool to support feature enhancement and user experience improvements in Indonesian mobile banking services.