<![CDATA[This study presents a comparative analysis of three transformer-based models, BERT, RoBERTa, and DistilBERT, for emotion classification on Indonesian text. The Indo4B dataset, which consists of five emotion labels (anger, fear, happiness, love, and sadness), is used as the benchmark corpus. All three models were trained using fine-tuning and hyperparameter-tuning procedures. The evaluation was conducted using accuracy, F1-score, model size, and training time as the primary performance metrics. The results indicate that RoBERTa achieves the best overall performance, with an accuracy of 90.83% and an F1-score of 91. Meanwhile, DistilBERT demonstrates approximately 60% faster training time compared to BERT, with only a 0.61% decrease in accuracy. These findings suggest that DistilBERT offers a highly efficient alternative while maintaining competitive classification performance.]]>
Copyrights © 2026
