<![CDATA[The Free Nutritious Meal (MBG) Program is one of the Indonesian government's public policies that has generated diverse public responses on social media, making sentiment analysis essential for understanding public perceptions as a basis for policy evaluation. This study aims to analyze public sentiment toward the MBG Program on the X social media platform using the IndoBERT model combined with the Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. A total of 2,260 tweets were collected through web scraping and subsequently preprocessed through text cleaning, normalization, tokenization, sentiment labeling using the InSet Lexicon, data balancing with SMOTE, and IndoBERT fine-tuning. Model performance was evaluated using accuracy, precision, recall, and F1-score metrics. The results show that the proposed model achieved an 80% accuracy, with approximately 75% of the data classified as negative sentiment, primarily driven by concerns regarding the state budget, potential corruption, and program distribution. Meanwhile, positive sentiment mainly highlighted the program's benefits for children's nutrition and local economic empowerment. These findings demonstrate that the combination of IndoBERT and SMOTE is effective for public policy sentiment analysis and can provide valuable insights for evaluating the implementation of the MBG Program. Keywords: Sentiment Analysis; Free Nutritious Food; IndoBERT; SMOTE; FFNN Abstrak Program Makan Bergizi Gratis (MBG) menjadi salah satu kebijakan pemerintah yang memunculkan beragam respons masyarakat di media sosial, sehingga diperlukan analisis sentimen untuk memahami persepsi publik sebagai bahan evaluasi kebijakan. Penelitian ini bertujuan untuk menganalisis sentimen masyarakat terhadap Program MBG di media sosial X menggunakan model IndoBERT dengan penerapan Synthetic Minority Over-sampling Technique (SMOTE) untuk mengatasi ketidakseimbangan kelas. Data dikumpulkan melalui proses scraping dan menghasilkan 2.260 tweet, kemudian diproses melalui tahapan pembersihan teks, normalisasi, tokenisasi, pelabelan sentimen menggunakan InSet Lexicon, penyeimbangan data dengan SMOTE, serta fine-tuning model IndoBERT. Evaluasi dilakukan menggunakan metrik akurasi, presisi, recall, dan F1-score. Hasil penelitian menunjukkan bahwa model memperoleh akurasi sebesar 80%, dengan dominasi sentimen negatif sekitar 75% yang dipengaruhi oleh isu anggaran, potensi korupsi, dan distribusi program, sedangkan sentimen positif menyoroti manfaat program terhadap gizi anak dan pemberdayaan ekonomi lokal. Temuan ini menunjukkan bahwa kombinasi IndoBERT dan SMOTE efektif untuk analisis sentimen kebijakan publik serta dapat menjadi masukan bagi evaluasi pelaksanaan Program MBG. Kata kunci: Analisis Sentimen; IndoBERT; Makan Bergizi Gratis; FFNN; SMOTE.]]>
