Recommendation System with Faster R-CNN for Detecting Content Violation in Broadcasting Videos

Television plays a significant role in achieving sustainable development as it can influence behavior and foster violence, hence why setting rules and standards is vital to ensure the well-being of vulnerable populations. Organizations such as Indonesian Broadcasting Commission use the Broadcasting Code of Conduct and Broadcast Program Standards and its five categories (5S) to regulate and monitorbroadcast content in Indonesia. With the growth of content and the urgency to monitor effectively by collaborating Artificial Intelligence with human reviewers, our research aims to develop a recommendation system to detect broadcast violations using Faster R-CNN. Data was collected using scraping based on the categories in the code of conduct and annotated manually. We have conducted four scenarios to train for three categories: SARU, SADIS, and SIHIR, using the Faster-RCNN ResNet-152 pre-trained model from the Tensorflow Object Detection API framework. The models are evaluated using mean Average Precision (mAP). The optimum results obtained are models with reduced learning rates with mAP scores of 0,325 for SARU, 0,2877 for SADIS, and 0,3563 for SIHIR. The models are then integrated into thesystem in which the back end is developed using the Flask framework and front-end using NextJS. Functionality and usability testing are carried out in which the system has met the requirement specifications of 100% and is easy to use with an average score of 6.75.