<![CDATA[General Background Social media comments offer valuable data for analyzing public discourse on policy issues. Specific Background This study investigates YouTube comments about Danantara, Indonesia's strategic investment body, using Natural Language Processing with 7,294 comments. Knowledge Gap Previous studies often analyze sentiment and topics separately, without integrated analysis or iterative labeling. Aims The study aims to classify sentiment using Support Vector Machine (SVM) and identify topics with Latent Dirichlet Allocation (LDA). Results 74.9% accuracy was achieved with SVM, classifying 58.0% of comments as negative, 29.3% neutral, and 12.8% positive. LDA revealed 6 topics for neutral, 4 for positive, and 3 for negative sentiment, with key concerns about transparency and corruption. Novelty This study integrates SVM and LDA with Human in the Loop labeling to capture both sentiment and topic substance. Implications Findings offer insights for improving transparency and public communication, while contributing to text mining in digital discourse. Highlights • The classifier achieved 74.9% accuracy after Human in the Loop labeling and manual verification.• Unfavorable polarity reached 58.0%, followed by neutral at 29.3% and positive at 12.8%.• Coherence scores selected 6 neutral, 4 positive, and 3 critical thematic clusters. Keywords Danantara; Sentiment Analysis; Topic Modeling; Support Vector Machine; Latent Dirichlet Allocation ]]>
Copyrights © 2026
