Inappropriate Content Classification based on Video Rating: A Preliminary Study

With the rise of social media, video analysis becomes more important in understanding user behavior. Video analysis has been a challenging problem in the computer vision domain compared to image analysis, considering the number of image frames per second a video can have, and the time-series information that underlying within the video. While recent advancements in video analysis improved gradually, the inappropriate content classification remains challenging, mainly due to the limited number of frames to be detected that is contributing to the inappropriate context of the video. In this work, we tackled this problem by addressing the current algorithm in video analysis problems and investigating the majority algorithm that is widely used in video analysis problems and is heavily affected by the issues in inappropriate content classification. A weighted score strategy is proposed to overcome the limitation of the small number of image frame detection in terms of inappropriate content so that inappropriate content classification problems can be optimized. To prove the concept, a video dataset labeled with inappropriate content inspired by Malaysia movie rating has been collected. The proposed strategy can outperform the majority algorithm in a good margin, showing that the proposed solution helps get more effective in inappropriate content classification.