<![CDATA[This study aims to compare the Inset Lexicon and MBERT sentiment labeling methods for analyzing sentiment related to the Danantara issue using a Support Vector Machine (SVM) with Linear, RBF, Polynomial, and Sigmoid kernels. The main issues in this study are the suboptimal sentiment labeling methods for Indonesian-language data that can accurately capture linguistic context, as well as the uncertainty regarding the best labeling method when combined with various TF-IDF-based SVM kernels. Model evaluation uses metrics such as Accuracy, Precision, Recall, F1-score, and Cross-validation (CV). The results show that the Inset Lexicon labeling method with a Linear kernel yields the highest Accuracy of 81% with a CV of 78%, Precision of 87%, and Recall of 92%. The RBF kernel achieved an Accuracy of 78% with a CV of 76%, followed by the Sigmoid kernel at 79% Accuracy with a CV of 76%, and the Polynomial kernel at 65% Accuracy with a CV of 65%. The highest F1-score for the negative class using the Linear kernel reached 89%. Meanwhile, in MBERT labeling, the highest accuracy was achieved by the RBF kernel at 72% with a CV of 71%, followed by the Linear kernel with an accuracy of 71%, then the Polynomial kernel with an accuracy of 68%, and the Sigmoid kernel with an accuracy of 67%. The highest F1-score was found in the negative class at 79% using the RBF kernel. Overall, the negative class showed the most consistent performance, while the neutral class had the lowest Recall and F1-score values for almost all kernel types. These findings confirm that an in-depth comparative analysis between lexicon-based and deep learning-based approaches demonstrates that methods such as the Inset Lexicon can deliver better and more stable performance on Indonesian-language data.]]>
