Sentiment Analysis of the Free Nutritious Meal Program on Twitter Using Naive Bayes and IndoBERT-Based Labeling

The Free Nutritious Meal Program is a government initiative aimed at improving the nutritional status of primary school children in Indonesia. However, its implementation has generated diverse public reactions on the X/Twitter platform, making systematic sentiment analysis essential for policy evaluation. This study analyzes public sentiment using two labeling approaches—translation-based TextBlob and IndoBERT contextual labeling—combined with Naïve Bayes and Linear SVC classifiers. A total of 2,903 Indonesian-language tweets were collected, preprocessed, and classified to compare the performance impact of each labeling method. The evaluation was conducted using accuracy, precision, recall, and macro F1-score. Sentiment distribution under IndoBERT indicates a predominance of negative and neutral opinions, particularly related to budget concerns, implementation quality, and food distribution issues. This study is subject to several limitations. The dataset size (2,903 tweets) and restricted temporal window may limit the generalizability of findings to long-term public discourse. The analysis also relies on a single social media platform (X/Twitter), excluding perspectives from other platforms such as Instagram or TikTok. Moreover, although IndoBERT improves contextual understanding, transformer-based labeling still may not fully capture sarcasm or highly colloquial expressions. Despite these limitations, the study demonstrates the effectiveness of combining Indonesian transformer models with conventional classifiers to support data-driven policy evaluation.