Utilizing Sentiment Insights for Software Evalution with Aspect-Based Sentiment Analysis in App Reviews Using IndoBERT-Lite

With mobile applications increasingly used daily, understanding user feedback is vital for software improvement. Manual analysis of user reviews on platforms is time-consuming and lacks scalability. This study utilizes aspect-based sentiment analysis with the IndoBERT-Lite model to automate the categorization and sentiment detection of feedback. Despite advancements, precise aspect classification and sentiment detection in Indonesian language reviews remain challenging due to limited resources. Using 5,000 Tokopedia reviews, labeled across five aspects, namely functionality, service, performance, usability, and design, the model achieved optimal results with batch size 32, showing 0.85 accuracy and 0.866 F1 score for aspect classification, and 0.998 across all metrics for sentiment classification.