A tensor-based method for learning ranking functions

The learning of ranking functions has recently gained much attention, and many methods based on SVM and Ensemble approaches have been proposed. Based on the above research, most of existing ranking learning methods take vectors as their input data, and then a function is learned in such a vector space for classification, clustering, or dimensionality reduction. However, in some cases, there are some reasons to take tensors as their input data, e.g., an image can be considered as a second order tensor. It is reasonable to consider that pixels close to each other are correlated to some extent. In this paper, we represent the data points by second order tensors rather than vectors, and then establish a new ranking learning model called Ranking Support Tensor Machine (RSTM), which based on support tensor machine. To solve this model, an iterative algorithm is used. This tensor-based algorithm requires a smaller set of decision variables as compared to vector-based approaches, thus, it is helpful to overcome small-sample-size problems in vector-based learning. We compare our proposed method with Ranking SVM on three databases. Experimental results show the effectiveness of our method. 1553-9105/Copyright © 2014 Binary Information Press.