Online gambling promotions on social media platforms such as YouTube often employ slang or non-standard language to evade traditional moderation systems, increasing the spread of illegal content in Indonesia. This study aims to develop a hybrid deep learning model combining BERT and LSTM to accurately detect online gambling promotions. Data were collected from YouTube comments through a web scraping process, followed by text cleaning, labeling, and normalization using a semi-automatic slang dictionary. The model was trained with slang-aware embeddings to capture informal language context. Evaluation was conducted using precision, recall, F1-score, and confusion matrix metrics. The results show an accuracy of 96% with an F1-score of 0.96, indicating a strong balance between precision and recall. These findings demonstrate the effectiveness of the proposed hybrid approach in automatically detecting online gambling promotional content.Kata kunci: Online Gambling Detection; Deep Learning; NLP; Slang-Aware Embeddings; BERT-LTSM AbstrakPromosi judi daring di media sosial seperti YouTube sering menggunakan bahasa tidak baku atau slang untuk menghindari deteksi sistem moderasi. Kondisi ini berpotensi meningkatkan penyebaran konten ilegal di Indonesia. Penelitian ini bertujuan mengembangkan model deep learning hibrida yang mengombinasikan BERT dan LSTM guna mendeteksi promosi judi daring secara lebih akurat. Data dikumpulkan dari komentar YouTube melalui proses web scraping, kemudian diproses melalui tahap pembersihan teks, pelabelan, dan normalisasi menggunakan kamus slang semi-otomatis. Model dilatih dengan slang-aware embeddings untuk menangkap konteks bahasa tidak resmi. Pengujian dilakukan menggunakan metrik precision, recall, F1-score, dan confusion matrix. Hasil menunjukkan akurasi sebesar 96% dengan nilai F1-score 0,96, menandakan keseimbangan tinggi antara presisi dan sensitivitas model. Temuan ini membuktikan efektivitas pendekatan hibrida dalam mendeteksi konten promosi judi daring secara otomatis.