Decoupled Collaborative Ranking

We propose a new pointwise collaborative ranking approach for recommender systems, which focuses on improving ranking performance at the top of recommended list. Our approach is different from common pointwise methods in that we consider user ratings as ordinal rather than viewing them as real values or categorical labels. In addition, positively rated items (higher rating scores) are emphasized more in our method in order to improve the performance at the top of recommended list. In our method, user ratings are modeled based on an ordinal classification framework, which is made up of a sequence of binary classification problems in which one discriminates between ratings no less than a specific ordinal category c and ratings below that category ({≥ c} vs. {< c}). The results are used subsequently to generate a ranking score that puts higher weights on the output of those binary classification problems concerning high values of c so as to improve the ranking performance at the top of list. As our method crucially builds on a decomposition into binary classification problems, we call our proposed method as Decoupled Collaborative Ranking (DCR). As an extension, we impose pairwise learning on DCR, which yields further improvement with regard to the ranking performance of the proposed method. We demonstrate through extensive experiments on benchmark datasets that our method outperforms many considered state-of-the-art collaborative ranking algorithms in terms of the NDCG metric.