Detecting Online Gambling Promotion in Youtube Comments Using Transformer-Based Models

YouTube comments section is often filled with online gambling promotional comments that are completely irrelevant to the video content. These comments typically share repetitive or templated patterns, making YouTube a major medium for online gambling promotion that negatively impacts social, economic, and psychological aspects of society. Transformer-based language models, particularly BERT, have shown strong performance in understanding contextual and semantic patterns in text. This study introduces the IndoGambling-Comments Dataset, a new Indonesian-language dataset containing labeled YouTube comments related to online gambling. Using this dataset, we compare the performance of several transformer models: IndoBERT, IndoBERTweet, RoBERTa, and XLM-RoBERTa. Experimental results show that IndoBERT achieves the best overall performance with an F1-score of 0.9850 when font preprocessing is applied. The inclusion of preprocessing consistently improves performance. Overall, the results demonstrate that transformer-based models are highly effective for online gambling comment detection, particularly when preprocessing strategies are aligned with model pretraining characteristics.