Ensemble Ranking SVM for learning to rank

This paper deals with the problem of learning to rank documents for information retrieval. Until now, Ranking SVM has been successfully used for learning to rank documents. The basic idea of Ranking SVM is to formalize learning to rank as a problem of binary classification on instance pairs and solve the problem using SVM. Even if Ranking SVM has achieved good ranking performances, there are some problems that its training time of train data sets grows exponentially when the size of the training set is large. In this paper, we propose a new method of learning to rank, named Ensemble Ranking SVM, which greatly improves the efficiency of the model training and achieves high ranking accuracy as well. In Ensemble Ranking SVM, each query of training sets is used to train a model using ensemble methods. Experimental results show that the performance of Ensemble Ranking SVM is quite impressive from the viewpoints of the accuracy and efficiency.