Analisis Sentimen Berbasis Aspek pada Ulasan Aplikasi JConnect Mobile Menggunakan IndoBERT dan mBERT

Abstract

JConnect Mobile is a mobile banking application developed by Bank Jatim that provides various digital banking transaction services. Although the number of users has continued to increase, the application still receives numerous complaints regarding its interface, features and performance, and service quality, as reflected in user reviews and its relatively low application rating. This study aims to compare the performance of the IndoBERT and mBERT models for aspect-based sentiment classification of JConnect Mobile user reviews. The dataset consists of 4,344 user reviews collected from Google Play Store and App Store between 2021 and 2025. The research methodology includes data collection, data labeling, preprocessing, feature extraction, modeling, and evaluation. The experiments were conducted using 5-fold cross-validation with 24 experimental scenarios involving combinations of hyperparameters, including 5 and 10 epochs, dropout rates of 0.2, 0.4, and 0.6, and batch sizes of 16 and 32. The results of the 5-fold cross-validation show that IndoBERT achieved the best performance, with mean F1-scores of 0.86 for the interface aspect, 0.83 for the features and performance aspect, and 0.74 for the service aspect. Based on these results, IndoBERT was selected as the best-performing model because it consistently outperformed mBERT across all aspects. Evaluation on the test dataset further demonstrated that the best-performing model achieved F1-scores of 0.80 for the interface aspect, 0.83 for the features and performance aspect, and 0.81 for the service aspect.