Implementasi Metode IndoBERT Untuk Analisis Emosi Pada Ulasan Pengguna Aplikasi M-Paspor

Abstract

The M-Paspor application serves as an essential digital platform for managing passport services online, making user reviews on the Google Play Store a valuable source for understanding public experiences with the system. These reviews contain various emotional expressions, ranging from appreciation to technical complaints, which require an artificial intelligence approach for effective analysis. This study employs the IndoBERT model due to its strong capability in interpreting Indonesian linguistic context through Deep learning methods. The research process includes collecting 2,996 user reviews, performing text cleaning and normalization, assigning initial sentiment labels using the Inset lexicon followed by manual verification, splitting the dataset, and fine-tuning IndoBERT with a maximum input length of 63 tokens. Evaluation on 446 test samples shows that IndoBERT achieves an accuracy of 86% and a macro F1-score of 0.61, demonstrating strong performance on positive and negative sentiment classes. This study concludes that IndoBERT is effective in identifying user emotions, although the neutral class remains challenging due to its limited representation and more ambiguous linguistic patterns. The findings also indicate inconsistencies between star ratings and textual sentiment in several reviews, suggesting that text-based analysis provides a more representative understanding of user perceptions toward the M-Paspor service.