Analisis Sentimen Berbasis Aspek pada Ulasan Aplikasi Layanan Publik

Studi Kasus Aplikasi MyPertamina Menggunakan Pendekatan Indobert dan Explainable AI (XAI)

Abstract

Application quality assessment on Google Play Store is generally based on numerical ratings, but this approach does not always reflect the actual sentiment of users. The difference between ratings and textual reviews has the potential to cause bias in application quality evaluation, especially for public service applications. This study aims to analyze the sentiment of MyPertamina app users in greater depth using an IndoBERT-based Aspect-Based Sentiment Analysis (ABSA) approach and to evaluate the correspondence between text sentiment and user numerical ratings. The research dataset consists of 9947 user reviews that have undergone preprocessing and aspect separation, with a focus on the account and payment aspects. The IndoBERT model was fine-tuned for sentiment classification and applied in an ensemble scheme to improve prediction stability. In addition, the Explainable Artificial Intelligence (XAI) approach using Integrated Gradients from Captum was used to provide interpretations of the model's prediction results. The results of the study show the dominance of negative sentiment in both main aspects, as well as the discovery of inconsistencies between textual sentiment and numerical ratings in some user reviews. These findings indicate that numerical ratings do not fully represent the actual user experience. Thus, this study contributes to the development of a more transparent and accurate aspect-based sentiment analysis, and offers a more comprehensive evaluation approach for improving the quality of digital public services.