A new recommender system framework for TV video

In recent years, recommender system has been one of the hottest topics in both industry and academia. The ability of providing personalized user experience has been a pivotal evaluation criterion to determine weather a company can be successful or not. In this paper, we fetch the first hand data from a Video Communication Co Ltd, and put forward a new recommender system framework for TV video. Based on Manhattan distance and considering programs popularity, we propose a new clustering algorithm, thus getting latent factors. According to similarity, we successfully map program to latent vector. Finally, we get the preference matrix by gradient descent method. Compared with SVDFeature on precision, recall and F1, it performs better.