<![CDATA[Detecting hate speech on Indonesian social media is challenging due to slang, abbreviations, and informal expressions that hinder automated text understanding. Traditional machine learning approaches often fail to capture contextual meaning effectively. This study aims to develop a hate speech detection system for Indonesian slang by evaluating contextual embedding IndoBERTweet combined with a Convolutional Neural Network (CNN) architecture. The research compares the performance of CNN and BiLSTM models using IndoBERTweet and FastText embeddings. A dataset of 1,477 labeled tweets categorized as Hate Speech, Abusive, or Non-Hate Speech was used. Evaluation metrics employed in this study consist of accuracy, precision, recall, F1 score, and AUC ROC. The results show that the IndoBERTweet + CNN model achieves the best performance, with 91.2% accuracy and a 91.1% F1-score, significantly outperforming FastText-based models. IndoBERTweet’s contextual embedding proves effective in handling the linguistic complexity and implicit meanings commonly found in Indonesian slang. These findings highlight the model’s strong capability for robust hate speech detection and open opportunities for its adoption as an automated content-moderation module that identifies and filters toxic narratives on social media platforms.]]>
Copyrights © 2026
