Public Sentiment on Indonesia’s Free Nutritious Meal Program: A Mixed-Methods NLP Evaluation

Large-scale nutrition intervention programs such as the Free Nutritious Meal Program (MBG) are likely to attract considerable attention on social media. While conventional evaluation techniques are often too slow to capture rapidly shifting sentiment, this study seeks to determine how sentiment can be evaluated. More specifically, we aimed to identify the key emerging issues. Methodology: In this study, one approach to examining emerging issues is to use a two-stage workflow in Natural Language Processing (NLP). The first step in sentiment analysis is using a transformer model (Indo-RoBERTa) to assign 'Positive', 'Negative', or 'Neutral' to 3,459 public texts from X (Twitter) social media. Secondly, we focused on 1,130 'Negative' texts. We used topic modeling (BERTopic) on this and identified the most critical clusters of issues to map and their relative importance. Results & Conclusions: Negative sentiment involves multiple factors, to which our model successfully highlighted four of the most impactful areas: (1) Financial concerns and budgetary priorities; (2) Responses to particular media coverage (e.g., Kompas); (3) Political general discourse; and (4) Expectations of particular local issues (education issues in Papua). Conclusion: Compared with the gaps in the program's nutrition components, the economic consequences, budget gaps, inequities, and regional policy deficiencies drew more public interest. Implications: The findings point to a clear need for a differentiated and open approach to communicating public policy. This approach should communicate the nutritional value and the need to align messaging with the public for the geographic and budgetary realities.