Recommending new movies: even a few ratings are more valuable than metadata

In the Netflix Prize competition many new collaborative filtering (CF) approaches emerged, which are excellent in optimizing the RMSE of the predictions. Matrix factorization (MF) based CF approaches assign low dimensional feature vectors to users and items. We link CF and content based filtering (CBF) by finding a linear transformation that transforms user or item descriptions so that they are as close as possible to the feature vectors generated by MF for CF. We propose methods for explicit feedback that are able to handle 60 000 features when the average number of non-zero features per item (or per user) is small. We collected movie metadata for Netflix Prize movies to conduct experiments with CBF on that dataset. We show that the prediction performance of the methods is favorable compared to that of CF, while their running time range between 10 minutes and an hour. We also evaluate some of the proposed algorithms on predicting ratings of new movies. We conclude that even 10 r...

• Type: Conference
• Year: 2009
• Where: RECSYS
• Authors: István Pilászy, Domonkos Tikk
• Keywords: Control Systems, Feature Vectors, Matrix Factorization, Netflix Prize, RECSYS 2009