<![CDATA[This study analyzes sentiment and conversational themes in 45 comments from 7 YouTube videos on Indonesia’s MBKM program. Sentiment classification is performed using an Indonesian pretrained model. Topic modeling is conducted with BERTopic, which leverages Transformer embeddings (paraphrase-multilingual-MiniLM-L12-v2), UMAP for dimensionality reduction, HDBSCAN for clustering, and c-TF-IDF for keyword extraction. The results indicate that most comments are positive (?93.3%), with relatively small neutral and negative portions. The very short and often formulaic nature of the comments limits lexical diversity, making topic differentiation challenging. The study’s contributions include a reproducible analytic protocol, c-TF-IDF–based topic reports, and methodological notes for modeling topics in very short texts. Future work should expand the corpus across channels and time periods and incorporate manual validation]]>
Copyrights © 2025
