<![CDATA[The rapid expansion of digital journalism in Indonesia has produced a large volume of long-form news articles covering complex socio-political and economic issues. One of the most widely discussed initiatives is the Free Nutritious Meal Program (Makan Bergizi Gratis, MBG), which has been reported from perspectives of nutrition, education, public policy, and economic sustainability. Due to the scale of coverage, manual analysis is impractical, requiring automated methods to extract latent themes. This study applies and compares three classical topic modeling techniques: Latent Dirichlet Allocation (LDA), Latent Semantic Analysis (LSA), and Non-Negative Matrix Factorization (NMF). The corpus was preprocessed using tokenization, stop-word removal, stemming, and normalization, followed by experiments with different numbers of topics. Model performance was evaluated with topic coherence and qualitative interpretability. The findings show that LDA provides the highest coherence and generates semantically rich topics. NMF yields sharper topic boundaries but sometimes lacks interpretability, while LSA demonstrates computational efficiency with lower semantic depth. Thematic analysis identified five dominant themes: policy implementation, nutritional adequacy, funding challenges, public responses, and governance risks. Temporal analysis revealed that MBG discourse has been present since 2020, but increased sharply in 2024 during Prabowo Subianto’s presidential campaign and grew further in 2025 with nationwide implementation. These results confirm that MBG is not only a nutrition and welfare policy but also a politically salient initiative intersecting with electoral strategies, public health, and socio-economic development.]]>
Copyrights © 2026
