Detecting Indonesian Online Gambling Promotions in Digital Images: An OCR-LSTM Pipeline

Abstract

The rapid spread of online gambling promotions via digital images on social media renders manual identification processes highly inefficient, necessitating an automated detection system. This study aims to implement a system to identify online gambling promotions in digital images using a combination of Optical Character Recognition (OCR) and Long Short-Term Memory (LSTM) methods. The research utilizes a dataset of 2,200 images, evenly balanced between gambling and non-gambling categories. The proposed system involves several key stages: image preprocessing using the Contrast Limited Adaptive Histogram Equalization (CLAHE) method to improve contrast, text extraction via PaddleOCR, text and vocabulary correction using FastText, and finally, contextual classification using the LSTM architecture. The testing results on 330 test data demonstrated excellent and stable model performance in recognizing text patterns. The model successfully achieved an impressive accuracy rate of 94%, with a precision of 0.96, a recall of 0.92, and an F1-score of 0.94. In conclusion, the combination of OCR and LSTM technologies is proven to be highly effective for automatically detecting online gambling promotional content.