<![CDATA[Abstract. Amid the wave of digital transformation, the Digital Korlantas Polri application emerged as a solution and breakthrough in the digital-based SIM (driver's license) issuance process. However, concerns regarding the application’s performance quality remain, one of which can be assessed through user comments on the Google Play Store. A disparity was found between the app’s rating and the content of user reviews, raising questions about the actual quality of the application. This study aims to identify the root causes of user issues with the Digital Korlantas Polri application and generate improvement recommendations based on the identified problems. The research utilizes the pre-trained IndoBERT model for sentiment classification, followed by semi-supervised topic modeling using Guided LDA to uncover hidden patterns in the review data. Furthermore, the pre-trained GPT-2 model is employed as a text generator to produce application improvement recommendations based on the identified issues. Evaluation results show that the sentiment model achieved a confidence score of 0.99, the Guided LDA model reached a coherence score of 0.51, and the GPT-2 model yielded a perplexity value of 1.3. Overall, the models successfully fulfilled their respective roles, enabling more effective and efficient analysis, and generating realistic and timely recommendations for addressing the identified issues]]>
Copyrights © 2026
