The CF+TF-IDF TV-program recommendation

Abstract

This paper first analyzes and discusses the traditional methods used in the TV-program recommendation system. Current studies show that the explicit data methods (user interest preference matrix) used in the real-world datasets does not work well and the precision of implicit data method (collaborative filtering) greatly relies on the data amount of each user. Then, we introduce a new method called CF+TF-IDF, which combines the collaborative filtering and TF-IDF algorithm. In order to analyze users’ preference, we also add k-means++ algorithm in it to cluster the users. The core of the method is to infer users’ preference from their viewing habits and the program type they choose. By using CF+TF-IDF, we build a TV-program recommendation model, aiming at improving users’ viewing experience.