Neighborhood preserving ordinal regression

Ordinal regression, which aims at determining the rating of a data item on a discrete rating scale, is an important research topic in pattern mining and multimedia data analysis. Most of the existing approaches of ordinal regression try to seek only one direction on which the projected data are well ranked. This setting largely limits the discriminative ability and may not describe the complicated distribution of the dataset very well. In this paper, we proposed a new algorithm called Neighborhood Preserving Ordinal Regression (NPOR), which aims to extract multiple projection directions from the original dataset according to the maximum margin and manifold preserving criteria. By optimizing the order information of the observations and preserving the intrinsic geometry of the dataset in a unified framework, NPOR provides faithful ordinal regression results to the new coming data samples. Experiments on various data sets demonstrate the effectiveness of the proposed algorithm.