<![CDATA[Social media now plays a crucial role as a platform for the public to voice opinions and views on public issues, including political policies such as the Draft Law on Regional Head Elections (RUU Pilkada). This study analyzes the performance of the Naïve Bayes, SVM, and KNN algorithms in classifying public sentiment towards the 2024 Regional Election Bill (RUU Pilkada) using two data labeling methods: manual and lexicon-based labeling. A total of 15,202 Instagram comments from Tempo.co, Kompas.com, and TvOnenews were analyzed, with 80% used as training data and 20% as testing data. The results show that SVM demonstrated the most consistent performance across both methods, achieving the highest accuracy of 78%. Naïve Bayes exhibited sensitivity to the labeling method, achieving 74% accuracy with lexicon-based labeling and 71% with manual labeling. KNN recorded the lowest performance, especially with lexicon-based labeling, achieving only 64% accuracy. The lexicon-based labeling method improved precision, recall, and F1-score across all algorithms, with SVM achieving the highest precision, recall, and F1-score at 76%, 76%, and 75%, respectively. This study highlights the significant impact of the labeling method on model performance, with lexicon-based labeling proving to be more effective in improving the quality of sentiment classification.]]>
Copyrights © 2025
