Topic-based sentiment analysis of Jamsostek mobile application reviews using the BERTopic method

Abstract

Rapid developments in digital technology have changed the way public services are delivered, shifting from conventional systems to mobile-based platforms. BPJS Ketenagakerjaan has adopted this transformation through the development of the Jamsostek Mobile (JMO) application. The application can be accessed on the Google Play Store, and users can provide ratings and reviews on the Google Play Store page. To improve the quality of the JMO application, it is important to pay attention to these reviews. Therefore, sentiment analysis is needed to identify and analyse positive and negative sentiments related to the use of the JMO application. The method used combines the IndoBERT and BERTopic models to understand user opinions in greater depth. This study shows that 75% of 2,846 comments contain negative sentiments, and the IndoBERT model achieves 100% accuracy in sentiment classification. In topic modelling, five clusters were formed, with the most discussed topics being requests for app updates and users' difficulties in accessing the JMO app due to constant errors and slowness.