<![CDATA[The rise of hate comments on social media, especially during politically sensitive periods such as Indonesia’s 2024 election has increased the urgency of automated cyberbullying detection. This study aims to evaluate and compare the performance of two Indonesian-language NLP models IndoBERT and Cendol in classifying hate speech on platform X (formerly Twitter). A total of 8,375 comments were collected and labeled into neutral, negative, and positive categories, with preprocessing steps including normalization, tokenization, and oversampling to balance class distribution. IndoBERT, trained on formal Indonesian corpora, achieved 90.7% accuracy and performed better on structured and formal texts. In contrast, Cendol, developed with informal social media data, scored 90.6% accuracy and showed superiority in identifying slang, sarcasm, and modified spellings. The findings highlight the complementary nature of both models. IndoBERT excels in recognizing policy-related or legal content, while Cendol is more effective in detecting casual hate speech. The study recommends ensemble learning strategies that integrate both models to improve content moderation systems in Indonesian digital platforms. These insights contribute to the development of more context-sensitive AI tools for hate speech detection in local languages. Keywords: hate speech; NLP; IndoBERT; Cendol]]>
Copyrights © 2025
