<![CDATA[Abstract — The Free Nutritious Meal Program (MBG) is a strategic policy of the Indonesian government to address stunting and improve human resource quality, yet its implementation has sparked public opinion polarization on social media X. This study aims to analyze public sentiment toward the MBG program to evaluate public responses and provide data-driven insights for policymakers. Traditional machine learning models often struggle to handle semantic complexity and the use of slang in Indonesian social media. Therefore, this study employs the state-of-the-art pre-trained IndoBERT model for sentiment classification. The research methodology involves a comprehensive pipeline implemented within a Graphical User Interface (GUI) application. A total of 3,459 tweets from X were collected, then processed through cleaning steps (removal of URLs, mentions, emojis), normalization using an Indonesian slang lexicon (233 entries), and sentiment labeling (positive, negative, neutral) with manual validation. The IndoBERT model was fine-tuned using 80% training data (2,366 tweets) with the following configuration: epochs = 5, batch size = 16, learning rate = 2e-5, and evaluated on 20% test data (592 tweets). The fine-tuning results achieved an accuracy of 87.16% and an F1-score of 87%, which is significantly higher than previous SVM-based methods (61–63%). Sentiment analysis on the entire cleaned dataset (2,958 tweets) showed a dominance of neutral sentiment (40.8%), followed by negative (30.6%) and positive (28.7%). Word cloud analysis revealed that public support focuses on the words "sehat" (healthy), "anak" (children), "gizi" (nutrition), while negative sentiment centers on "anggaran" (budget), "korupsi" (corruption), and "jakarta". This study concludes that fine-tuned IndoBERT is effective for public policy sentiment analysis on complex Indonesian social media texts, and provides recommendations for the government to improve budget transparency and public communication. The novelty of this study lies in the development of a GPU-accelerated end-to-end desktop application that integrates the entire Natural Language Processing pipeline, making it a practical real-time evaluation tool for non-technical government analysts. Keyword — IndoBERT, Natural Language Processing, Public Policy, Sentiment Analysis, Free Nutritious Meal Program]]>
