<![CDATA[This study proposes a hybrid approach to sentiment analysis by combining lexicon and Logistic Regression methods based on TF-IDF features, and using Multinomial Naive Bayes as a benchmark, to classify YouTube comments related to Suharto's appointment as a National Hero. The dataset contains 500 comments TV taken from five YouTube channels: TalkShow tvOne Reload, Kompas Madiun, Tribunnews, Perspektif, and Tempodotco, 100 comments each. The research stages include text pre-processing, lexicon development, automatic labeling, manual validation of annotations, TF-IDF feature extraction, and sentiment classification. The validation results show a 63% agreement between lexicon labels and manual annotations, with a Cohen's Kappa of 0.15 indicating low inter-annotation agreement. The proposed Logistic Regression framework achieved an overall accuracy of 91%, with a weighted average F1 score of 0.87 and a Macro F1 score of 0.43, slightly better than Naïve Bayes Multinomial, which achieved an accuracy of 90% and a weighted F1 score of 0.85. However, the relatively small performance improvement and high accuracy affected the dominance of the neutral class. The Confusion Matrix for the data shows that the model still has difficulty distinguishing between positive and negative comments the recall for the negative class is 0%, and for the positive class is 20%. Therefore, these results are still preliminary and cannot be generalized.]]>
Copyrights © 2026
