Semantic Video Recommendation System Based on Video Viewers Impression from Emotion Detection
Darari Nur Amali, Ali Ridho Barakbah, Adnan Rachmat Anom Besari, Dias Agata · 2018
Huge video data that available at many video hosting providers create a new problem. One of them is a problem in video discovery system. One of many solutions that solve the problem is video search function. But, video search function cannot provide best user experience despite using a good keyword. This problem encourages many video hosting service providers to get deeper insight of the video data that they have and used them to give the best experience to the user. A video service provider such as Youtube will perform video analysis based on the internal video contents such as colours, textures, shapes, music signal, video title provided by users and other features that exist in video. The result of this analysis was used to understand its video content and also user preference on a video. At the last, an insight about user personalities can be known and video provider can personalize video for each user to give them best user experience. With technological developments in Machine Learning and Computer Vision technology, video analysis can be done based on other things beyond the video. It was the audience's impression of a video. An analysis of audience impressions in real-time was expected to get insights of a video content by a real feeling of the viewers. By those insights, a new video recommendation system can be built to deliver a new experience for the user in video discovery system. This system is intended to be able to deliver variative videos which match their personal emotion at the time. Based on this research results, 65% users love more than half in video list that provided by impression-based recommender system.