Sentiment Analysis of Indonesian Netizens toward Vasectomy on X Using the IndoBERT Model

Vasectomy-related conversations on X (Twitter) frequently generate polarized pro–contra debates that can shape public understanding of male contraception, yet evidence on Indonesian netizens’ sentiment remains limited. This study maps and classifies sentiment toward vasectomy during April–June 2025 using a descriptive quantitative text-mining and NLP pipeline. After preprocessing (cleaning and deduplication), 9,817 posts were analyzed. Semi-supervised labeling was performed using the teacher model taufiqdp/indonesian-sentiment with confidence-based refinement, supported by a rule-based sarcasm_flag that identified 330 potentially sarcastic texts. A 20% manually verified GOLD subset (1,963 samples) served as ground truth, and IndoBERT (indolem/indobert-base-uncased) was fine-tuned with weighted cross-entropy and early stopping. Evaluation on the GOLD test set (n = 393) showed strong performance (accuracy = 0.8168; macro F1 = 0.8141), with most errors concentrated in short, ambiguous, or humor/sarcasm-leaning posts. Full-corpus predictions produced 3,957 negative, 3,520 positive, and 2,340 neutral texts, indicating a contested and polarized discourse with a slightly higher negative share. These findings support the need for evidence-based digital communication strategies to address misconceptions and stigma surrounding male contraception.