<![CDATA[User reviews on the Google Play Store can serve as an important data source for understanding public perceptions of mobile banking services. At the time this paper was written, the application’s average user review score was 3.2, indicating a poor rating. Therefore, the bank needs to further examine the factors that could improve this rating in order to minimize reputational risk, since most users who intend to install an application tend to check its rating first before deciding whether to install it. In general, a rating considered very good is at least 4.5. This study aims to analyze user review sentiment toward the Bank Kalbar Mobile Banking application using a natural language processing approach with IndoBERT, along with two comparison models: TF-IDF with Logistic Regression and RNN BiLSTM. The dataset consists of 2,465 reviews classified into three sentiment classes: positive, negative, and neutral. The data distribution shows class imbalance, with 1,445 positive reviews, 894 negative reviews, and 126 neutral reviews. The data were split using a stratified method into 70% training data and 30% testing data. The research stages included text cleaning, data splitting, model training, and evaluation using accuracy, macro-F1, precision, recall, and a confusion matrix. The experimental results show that the TF-IDF + Logistic Regression model achieved the best performance, with an accuracy of 0.8581 and a macro-F1 score of 0.6777. The RNN BiLSTM model obtained an accuracy of 0.8311 and a macro-F1 score of 0.6777, while IndoBERT achieved an accuracy of 0.8041 and a macro-F1 score of 0.6774. Although IndoBERT did not achieve the highest accuracy, it demonstrated better capability in identifying the neutral class, as indicated by a recall score of 0.6316. These findings indicate that, for a relatively small and imbalanced dataset, a classical TF-IDF-based model can still deliver competitive performance compared with deep learning and transformer-based models.]]>
Copyrights © 2026
