<![CDATA[Public opinion on social media is a crucial representation of government policy legitimacy, especially in the fiscal sector. This study intends to provide a comparative investigation of the efficacy of sentiment categorization on YouTube comments pertaining to the activities of the Indonesian Finance Minister by juxtaposing the Support Vector Machine (SVM) algorithm with the RoBERTa Transformer model. A total of 3,780 comments were acquired from national digital media channels. The research method involves intensive text preprocessing, including stemming using the Sastrawi algorithm and lexicon-based labeling. The results showed that the SVM algorithm with TF-IDF features achieved an accuracy of 83.33% and an F1-score of 76.05%. In contrast, the RoBERTa model showed a significantly lower performance with an accuracy of 29.76%. This study concludes that for datasets dominated by neutral sentiments and informal language in specific Indonesian contexts, traditional machine learning like SVM with optimal feature engineering remains more reliable and efficient than complex Transformer models that require more extensive fine-tuning.]]>
Copyrights © 2026
