Application of Ensemble Machine Learning Methods for Aspect-Based Sentiment Analysis on User Reviews of the Wondr by BNI App

Abstract

This study analyzes user perceptions of the Wondr by BNI app using Aspect-Based Sentiment Analysis (ABSA) and a stacking ensemble learning approach on user reviews. Data were collected from the Google Play Store and App Store through scraping, then processed and labeled. The study involves two classification stages: aspect identification and sentiment classification for each aspect. The stacking ensemble model without resampling showed the best performance, with F1-scores of 99.4% for UI (User Interface), 99.3% for Authentication, and 99% for Transaction. For sentiment classification, F1-scores reached 82.2% User Interface (UI), 87.8% (Authentication), and 92.4% (Transaction). The use of LIME (Local Interpretable Model-Agnostic Explanations) improved model interpretability by highlighting keywords influencing the classification results. The final output of this research is a website capable of performing aspect-based sentiment classification