<![CDATA[The Free Nutritious Meal Program has sparked debate on the X platform, creating the need to map public sentiment toward the policy. This study aims to map public sentiment toward the MBG policy on X and compare the performance of two Indonesian NLP models, IndoBERT and DistilBERT, under both imbalanced and balanced data conditions. The corpus consists of public tweets collected from 13 August to 17 September 2025, followed by text cleaning and automatic labeling. Class imbalance is addressed through back-translation to obtain more even class proportions. Four scenarios are evaluated using accuracy, precision, recall, and F1-score. On the original imbalanced data, IndoBERT reaches 96.32% accuracy with a macro F1 score of 0.9565. After balancing, interclass performance improves with a macro F1 score of 0.9384. DistilBERT remains competitive and more efficient, with accuracy around 91% to 93%. These findings underline the importance of aligning model choice and balancing strategy with analytic objectives.]]>
Copyrights © 2026
