The large volume of user reviews for Duolingo in Indonesia makes the manual evaluation process difficult to conduct. Therefore, this study aims to analyze user sentiment to understand their responses and experiences with the Duolingo application. This study utilizes the IndoBERT model to classify reviews into three sentiment categories: positive, neutral, and negative. The evaluation results demonstrate that the IndoBERT model performs exceptionally well, achieving an Accuracy rate of 92% and a Macro F1-Score of 71%. The majority of the analyzed reviews fall into the positive sentiment category (83.47%), indicating a high level of user satisfaction with the application. The positive sentiment class achieves an F1-Score of 97%, whereas the neutral class yields the lowest Macro F1-Score at 44% due to many neutral reviews being brief and ambiguous. The findings of this study show that IndoBERT is effective and reliable in automatically mapping public opinion and providing an overview of user perception toward the Duolingo application
Copyrights © 2026