Evaluating Public Trust in the Animation Industry: A Comparative Sentiment Analysis Using Random Forest and Fine-Tuned IndoBERT

Abstract

The release of local animated blockbusters in 2025, particularly "JUMBO," established a new benchmark for Indonesia's creative sector. However, public discourse has shifted from patriotic support to rigorous quality assessment. This study investigated public trust in the Indonesian animation industry through sentiment analysis by comparing a traditional machine learning approach and a deep learning model. We utilized Random Forest with TF-IDF as a baseline and a fully fine-tuned IndoBERT model as the proposed method, supported by a hybrid dataset strategy designed to address class imbalance. Experimental results showed that the Random Forest model achieved a high accuracy of 97.96% but struggled with ambiguous and context-dependent sentences due to its reliance on word frequency. In contrast, the fine-tuned IndoBERT model achieved 100% accuracy, precision, recall, and F1-score by effectively capturing semantic context, negation, and contrast through self-attention. Comparative analysis with recent studies confirmed that the proposed approach outperformed existing benchmarks, highlighting the effectiveness of balanced data construction and Transformer-based architectures for Indonesian sentiment analysis. Qualitative findings further revealed a shift in public sentiment from national pride toward more critical evaluations of narrative quality and voice acting, indicating a maturing audience. These results demonstrate that fine-tuned IndoBERT provides a robust and reliable framework for evaluating public trust and sentiment in the Indonesian animation industry.