The Transjakarta bus fare increase planned to IDR 5,000-7,000 triggered public debates on social media. The study aims to measure sentiment proportions and compare Naive Bayes and Support Vector Machine performances in classifying TikTok comments on the issue. Datas were collected through six news videos web scraping, yielding 2,820 comments, reduced to 1,000 samples through Simple Random Sampling and labeled by five annotators with a Krippendorff’s Alpha of 0.7827. After removed neutral class, 876 data remained (67.8% negative, 32.2% positive). Datas were processed using Sastrawi, extracted with TF-IDF, and modeled using 80:20 hold-out with Stratified 5-Fold Cross Validation for hyperparameter tuning. ComplementNB variant outperformed SVM, achieving 84% accuracy and 0.83 F1-Macro. Temporal analysis revealed strong correlations (r = 0.92, though based only three weekly data points) between time and intensifications of negative sentiments, while thematic explorations uncovered four public objections: subsidy inequity, modal-shift, payment system, and demands for service improvement.
Copyrights © 2026