<![CDATA[Automatic emotion analysis on informal Indonesian texts is a challenging task due to high linguistic variation, the use of slang, and abbreviations. This research focuses on the development and evaluation of an accurate emotion classification model, which can serve as a core component various relevant Natural Language Processing (NLP) applications. The proposed method is the fine-tuning of the pre-trained language model IndoBERT to classify texts from the social media platform Twitter (X) into five emotion classes: anger, fear, happy, love, and sadness. A custom dataset consisting of 4,940 Twitter posts was built through a targeted scraping process and statistically validated labeling to ensure data relevance and balance. Experiments show that after undergoing a comprehensive text preprocessing stage, including normalization using a custom abbreviation dictionary and stemming, the fine-tuned model achieved very high performance. Evaluation results on the test data show the model successfully reached an accuracy of 94% and a weighted average F1-score of 0.94. Learning curve analysis also confirms that the model did not suffer from overfitting and possesses good generalization capabilities. These results demonstrate that the IndoBERT fine-tuning approach is a highly effective and reliable solution for emotion classification in the informal Indonesian text domain.]]>
Copyrights © 2025
