Sentiment Analysis of Indonesia’s Free Nutritious Meal Program on Platform X (Formerly Twitter) Using IndoBERT

Abstract

Public sentiment toward government programs is increasingly expressed through social media, necessitating robust quantitative evaluation methods. This study examines public sentiment toward Indonesia’s Free Nutritious Meal (Makan Bergizi Gratis/MBG) program using 7,958 manually annotated Indonesian-language posts from platform X (January–August 2025), consisting of 3,752 positive, 848 negative, and 3,358 neutral tweets. Sentiment classification was conducted using IndoBERT-base-P2 and compared with a Support Vector Machine (SVM) baseline with TF-IDF features, employing class-weighted learning to address data imbalance. Model performance was evaluated using accuracy and macro F1-score, followed by paired-sample statistical testing. IndoBERT-base-P2 achieved 92% accuracy and a macro F1-score of 0.90, outperforming SVM (86% accuracy, macro F1 = 0.83). Paired t-test results indicate that this improvement is statistically significant (p < 0.05), confirming the robustness of transformer-based modeling. This study contributes methodologically by integrating contextual language modeling, imbalance-aware optimization, and inferential statistical validation within a unified sentiment analysis framework, demonstrating the quantitative advantage of transformer-based approaches for Indonesian social media policy analysis.