Penerapan Model Roberta Untuk Deteksi Emosi Multikelas Berbasis Teks Bahasa Inggris

Abstract

The rapid advancement of digital technologies has significantly increased the prevalence of text-based communication, particularly on social media platforms that serve as primary channels for emotional expression. In this context, emotion detection from text plays a crucial role in various applications, including opinion analysis, intelligent interaction systems, and psychological monitoring. This study implements the RoBERTa model to perform multi-class emotion classification on English-language text, targeting six primary emotion categories: sadness, joy, love, anger, fear, and surprise. The dataset used in this research comprises 20,000 text samples collected from public repositories, processed through comprehensive preprocessing and tokenization stages. The RoBERTa model is fine-tuned using an optimized set of hyperparameters obtained through systematic tuning, including adjustments to learning rate, batch size, epoch count, and weight decay. Model performance is evaluated using accuracy, precision, recall, and F1-score metrics on both training and testing data. Experimental results indicate that the model achieves high classification accuracy and generalization capability without signs of overfitting. These findings confirm that RoBERTa is an effective and robust approach for emotion detection tasks in text, with strong potential for broader deployment in natural language processing (NLP) applications powered by artificial intelligence.