On-line Weighted Matrix Factorization for TV Program Recommendation

Matrix Factorization (MF) is known as an effective technique in collaborative filtering for recommendation. The MF approaches have often been applied for movie recommender systems which have user rating data. However, they cannot effectively be applicable for TV program recommender systems because (i) explicit rating values are not available; (ii) many TV programs are broadcast under single TV program titles such as News, Shows, Dramas etc.; and (iii) the preferences of TV viewers on TV programs are subject to change in time. Therefore, in this paper, we propose an MF technique that considers such problems, thus making the MF technique suitably applicable for TV domain. We also present experimental results to show the effectiveness of our proposed extended MF technique.