Deteksi Ujaran Kebencian Bahasa Indonesia Menggunakan Fine-Tuning IndoBERT Berbasis Transformer

Abstract

Hate speech on Indonesian social media continues to increase, creating a need for automated systems capable of accurately classifying textual content. This study proposes hate speech detection using the pre-trained IndoBERT transformer model through fine-tuning. Unlike previous studies that combine transformers with additional architectures such as CNN or BiLSTM, this study directly evaluates IndoBERT without additional deep learning layers. The dataset consists of 13,169 Indonesian tweets annotated into Hate Speech (HS) and Non-Hate Speech (Non-HS) categories. The dataset was divided into 80% training, 10% validation, and 10% testing data. The model was trained for three epochs using the AdamW optimizer with a learning rate of 2e-5 and a batch size of 8. Experimental results show that IndoBERT achieved 90.43% accuracy, 90.11% precision, 90.36% recall, and a 90.23% F1-score. These results demonstrate that direct fine-tuning of IndoBERT can achieve strong classification performance with a simpler architecture, supporting its potential for automated digital content moderation.