A New Discriminative Ordinal Regression Method

Ordinal regression as an important machine learning problem has been widely applied to information retrieval and collaborative filtering. Current ordinal regression methods include perceptron based (Pranks) methods, SVM based ordinal regression (SVOR) methods, and discriminant learning ordinal regression (KDLOR) method etc. Among these methods, KDLOR performs well for rank prediction, because of its majority on discriminant information for classes. However, this method only considered the ordinal information of two adjacent classes for discriminant learning. In fact, there exists much ordinal information in any upper-lower pair class. In this paper we present an enhanced discriminant learning ordinal regression (EDLOR) method which uses global ordinal constraints on any upper-lower pair class, simultaneously realize the maximum distance of them. The results of numerical experiments on Synthetic and benchmark datasets verify the usefulness of our approach.