Learning Preferences with Co-Regularized Least-Squares

Situations when only a limited amount of labeled data and a large amount of unlabeled data is available to the learning algorithm are typical for many real-world problems. In this paper, we propose a semi-supervised preference learning algorithm that is based on the multiview approach. Multi-view learning algorithms operate by constructing a predictor for each view and by choosing such prediction hypotheses that minimize the disagreement among all of the predictors on the unlabeled data. Our algorithm, that we call Sparse Co-RankRLS, stems from the single-view preference learning algorithm RankRLS. It minimizes a least-squares approximation of the ranking error and is formulated within the co-regularization framework. The experiments demonstrate a signi cantly better performance of Sparse Co-RankRLS compared to the standard RankRLS algorithm. Moreover, our semi-supervised preference learning algorithm has a linear complexity in the number of unlabeled data items, making it applicable to large datasets.