Klasifikasi Emosi Multikelas Berbasis Teks Bahasa Indonesia Menggunakan Indoroberta

Abstract

The digital revolution has driven significant changes in human communication patterns around the world. By 2025, the number of global internet users is expected to reach 5.56 billion people, with more than 76.9% of the world’s population using social media. Social media has become the primary arena for emotional expression and the dissemination of opinions, particularly on platforms such as WhatsApp, Instagram, TikTok, and X (Twitter). However, the main challenge faced is the diversity of emotional expressions conveyed through text in various languages. IndoRoBERTa, as a transformer model adapted for the Indonesian language, is utilized in this study to perform multi-class emotion classification on Indonesian-language text, with emotion categories such as anger, fear, joy, love, sadness, and neutral. The research data is derived from a combination of two public datasets processed through comprehensive preprocessing steps, class balancing, and stratified data splitting to maintain the proportion of emotion labels. This study employs fine-tuning techniques on the IndoRoBERTa model and hyperparameter optimization using Optuna to find the optimal configuration, including batch size, learning rate, and dropout parameters. Model evaluation is conducted using accuracy, precision, recall, F1-score, and confusion matrix metrics, which indicate that the model achieves a macro F1-score of 0.70 on the test data. The model’s performance shows consistency in predicting majority emotions such as anger, fear, joy, and love, although challenges in distinguishing between neutral and sad classes remain. The results of this study contribute to the development of text-based emotion classification systems for the Indonesian language and can be applied to various NLP applications such as public opinion analysis, social media monitoring, and the development of intelligent chatbot systems.