A real-time video recommendation system for live programs

Few video recommendation systems could work in a live broadcast situation, partly because that the video-related information is hardly to be obtained when the live program is played for the first time. The lack of the video information, as well as the user profile, and the self-limitation of the recommendation algorithm lead to a disability for the current video hosting systems in providing a good recommendation. This paper presents a novel system to alleviate this problem. Given little information on the live program, our proposed system could find effective ways to collect the necessary information on both the live program and the subscribers and to recommend the relevant online videos to the subscribers. By utilizing user profiles and his/her social relationship, a user preference model has been constructed in our work, which greatly improves the diversity of the recommendation results. Then, we dispose some sort of new recommendation strategies in use at our demo system to meet each special need of diverse live programs and throw light upon how to fuse these strategies.