This study examines Aspect-Based Sentiment Analysis (ABSA) of public opinion regarding Danantara's policy on platform X, the primary space for public discourse on strategic economic policies in Indonesia. This research matters because conventional sentiment analysis has not revealed the policy dimensions that attract public attention. The research data consists of 14,123 Indonesian-language tweets from February 24–28, 2025. Aspect identification was conducted exploratively using Latent Dirichlet Allocation (LDA) and focused on three main aspects: policy transparency, management effectiveness, and institutional credibility. Sentiment classification at the aspect level was performed using the IndoBERT–BiLSTM model and compared with the fine-tuned IndoBERT model. The test results showed that IndoBERT–BiLSTM achieved an accuracy of 90.83% and an F1-score of 90.86%, higher than the fine-tuned IndoBERT with an accuracy of 89.51% and an F1-score of 89.43%, resulting in a performance increase of 1.32% and 1.43%. Aspect-by-aspect evaluation showed that IndoBERT–BiLSTM achieved 99% accuracy for policy transparency, 95% for management effectiveness, and 94% for institutional credibility, respectively. These findings confirm the effectiveness of the IndoBERT–BiLSTM integration for the Indonesian-language ABSA and are relevant as a baseline for a more targeted understanding of public perception.
Copyrights © 2026