A content and user-oblivious video-recommendation algorithm

Videos have become popular on Internet, and corresponding video-recommendation algorithms have become an important factor for maintaining user’s satisfactory level and the profits of video-service providers. Existing recommendation algorithms are often highly dependent on the precious information of contents and users. However in many scenarios those information is not available for privacy reasons. This paper provides a watching sequence-based video-recommendation algorithm that can work well without the video contents and users information. The algorithm consists of three models: the exactly prefix matching tree, partial prefix matching tree, and the postfix matching tree. The final recommendation results are composed from the three models. The corresponding search tree, the matching search tree, and weight calculating algorithm are developed for each model. The algorithm is evaluated based on the half-year log files of a practical video website. The experimental results show that our algorithm performs better on execution time, accuracy, diversity of recommendation results, and non-hot coverage than the traditional recommendation algorithms.