Deteksi Komentar Judi Online Menggunakan Bidirectional LSTM dan FastText pada Platform Video Daring

Abstract

Online gambling promotions are spreading widely in the comment sections of video platforms, and they are difficult to stop using traditional keyword-based filters. This is because promoters keep changing the way they write, using symbols, unusual spellings, or informal language to avoid being detected. This study aims to build a reliable system to detect these gambling-related comments using deep learning. The method combines Bidirectional Long Short-Term Memory (Bi-LSTM) and FastText word embedding to understand both the meaning and structure of the words in each comment. The dataset includes 47,737 comments in Indonesian from YouTube, divided into training (70%), validation (15%), and testing (15%) sets. After going through several text-cleaning steps, the model was trained and tested. The results were very promising: the model achieved 98.81% accuracy overall. For gambling-related comments, it reached 97.71% precision, 96.76% recall, and 97.24% F1-score. These results show that Bi-LSTM combined with FastText is an effective and powerful solution for automatically detecting online gambling content in video comments.