<![CDATA[Jakarta, the capital of Indonesia, faces persistent challenges with its public transportation system due to rapid urbanization, increased use of private vehicles, and poor service quality. While social media platforms such as X (formerly Twitter) offer valuable insights into public opinion, their unstructured nature complicates analysis. This study uses deep learning models to categorize user sentiments into six labels that cover positive and negative aspects of comfort, safety, and punctuality. The results show that IndoBERTweet achieved the highest performance, with 95.43% accuracy and a macro F1-score of 0.9545. It also required the shortest training time, at six minutes and 30 seconds. IndoBERTweet+GRU followed closely behind with an accuracy of 94.62% and a macro F1-score of 0.9460 in six minutes and 50 seconds. This shows that adding a GRU layer provides competitive results, but does not surpass the baseline model. Error analysis revealed that, while the models performed well with explicit sentiments, the models struggled with implicit expressions, such as sarcasm and mixed opinions. These results demonstrate the potential of sentiment analysis in real-time monitoring systems, which could help policymakers identify urgent issues and support data-driven improvements in Jakarta’s urban transportation services.]]>
Copyrights © 2025
