Detection of Online Gambling Promotional Comments on YouTube Using IndoBERT

Online gambling promotional comments on YouTube have become increasingly widespread and are difficult to detect with conventional moderation systems because they use informal language, varied spellings, and disguised keywords. This study applied the IndoBERT Transformer model to detect such comments using a cross-category video dataset in order to improve the generalization capability of the model. The research followed the Cross-Industry Standard Process for Machine Learning framework, which covers business and data understanding, data preparation, modelling, evaluation, deployment, and monitoring and maintenance. A total of 210,461 comments were collected through the YouTube Data API from 14 video categories. Preprocessing included emoji-to-text conversion, character normalization, text cleaning with regular expressions, and stopword removal using PySastrawi, NLTK, and a custom stopword list. The learning rate, batch size, and number of epochs were optimized using Optuna. On the test data, the model achieved an accuracy of 0.99591, a precision of 0.96396, a recall of 0.93265, and an F1-score of 0.94805. These results indicated that IndoBERT was effective in detecting online gambling promotional comments on YouTube and generalized well across various comment patterns.