Study And Implementation Of IPTV Recommendation System

With the development of our society,we can see Internet everywhere in our life.Traditional TV has been superseded by IPTV(Internet Protocol Television).IPTV has richer content and higher interactivity than traditional TV.But the richness of IPTV's programs makes it difficult for users to find interesting programs to watch.The application of recommendation system on IPTV makes it simpler for users to find those interesting programs.Generally,recommendation algorithms can be classified into three kinds,content-based recommendation,collaborative filtering and hybrid recommendation.In this paper,we propose an IPTV recommendation system combining content-based recommendation and collaborative filtering.IPTV's programs have many types and one user may have different tastes on different types of programs.If we leave out program types when making recommendation,there may be some influence on the accuracy of recommendation.In this paper,taking multi-type problem into consideration,we propose a content-based recommendation algorithm,Multi-type K-Nearest Neighbors(MKNN),based on K-Nearest Neighbors(KNN)and implement it with distributed and parallel approaches.As to collaborative filtering,we propose Multi-type Regularized Singular Value Decomposition(MRSVD)based on Regularized Singular Value Decomposition(RSVD).When combining content-based recommendation and collaborative filtering,we use a self-adaption Weighted algorithm based on Weighted algorithm using stochastic gradient descent.Experimental results on a real IPTV dataset from a provincial-level broadcasting corporation show that MKNN,MRSVD and self-adaption Weighted give more accurate recommendation results than KNN,RSVD and Weighted.