| dc.description.abstract | Online gambling has become a serious issue in Indonesia involving various layers of society with continuously increasing annual transaction volumes. The high usage of internet and social media in Indonesia, particularly platforms like Instagram through video reels content, has become a target for embedded online gambling advertisements (content marketing). This research aims to develop and design a website-based system that can detect the possibility of embedded online gambling advertisements in Instagram reels videos using the Faster R-CNN deep learning algorithm, by identifying the forms and patterns of advertisement appearances. The methodology used includes collecting reels video datasets (containing and not containing advertisements), data processing in the form of frame extraction into sample images and advertisement area annotation, AI model development using Python and PyTorch deep learning framework, as well as model performance evaluation and system testing. The research results show that the developed Faster R-CNN model has good detection capabilities on test data. The best model (learning rate 0.005) achieved mAP 0.7849, Precision 0.9900, Recall 0.8268, F1 Score 0.9011, and Mean IoU 0.8528. The implemented website-based system (Ada Judol) functions well, capable of receiving video upload input or URLs, and displaying detection result of video sample images with bounding boxes, timestamps, and detection confidence scores with an average response time of approximately 4.5 seconds. Although there are limitations in this research, such as detection limited to the first 15 seconds of video, inability to validate the accuracy of online gambling advertisements, not yet practically beneficial for Instagram or similar platforms, as well as limitations in certain advertisement backgrounds and positions, this research successfully realizes an automatic detection tool for embedded online gambling advertisements in Instagram reels videos. | en_US |
