Aspect-Based Sentiment Analysis of the BRImo Application Based on User Reviews on Google Playstore

Abstract

BRI is focused through their mobile banking superapp known as BRImo. However, based on discussions with the BRImo application research team, The user reviews available on the Google Play Store application service page could be classified based on the PACMAD usability aspects and sentiment using several classification models, including Random Forest, Decision Tree, and Extreme Gradient Boosting, with TF-IDF employed as the feature extraction method. Additionally, the Random Oversampling and Synthetic Minority Oversampling Technique (SMOTE) methods were applied as supplementary treatments to address the issue of imbalanced classes in BRImo application user review data. Topic modeling was also conducted using the LDA method to identify keywords and the main discussion topics for each PACMAD usability aspect and its sentiment, resulting in clear topics that can serve as a focus for the development of the BRImo application. The research findings indicate that the XGBoost classification model, combined with the SMOTE sampling method, demonstrated the best performance in classifying PACMAD usability aspects and sentiments, achieving F1-scores of 86.55% and 89.59%, respectively. Furthermore, the key topics for each PACMAD usability aspect and their associated sentiments were identified.