An Efficient Collaborative Recommendation Technique for IPTV Services

In this paper, we propose an efficient collaborative recommendation technique for IPTV services. Our technique solves the scalability and sparsity problems which the conventional algorithms suffer from in the IPTV environment characterized by frequent additions and deletions of contents and the large scale in the size of users (homes) and contents. The technique consists of two steps: user group profiling and user-based collaborative recommendation. For user group profiling we propose a scoring method to extract efficient user features from the TV watching history of users. We use CF-IUF, a modified TF-IDF, to represent user-based TV watching patterns by category through analysis of channels and menus of IPTV services. Users are grouped using the ISOData algorithm based on the feature vectors composed of the CF-IUF scores by category. The user-based collaborative recommendation is performed using the profile information of users similar to the target user. We use Pearson's correlation coefficient and Spearman's ranking coefficient to compare and select similar users. We experimented with our method using the data of the actual one-month IPTV services. The experiment results showed the success rate of 93.58% and the precision of 77.40%, which are considered a good performance for IPTV services.