ARSM Gradient Estimator for Supervised Learning to Rank

ABSTRACT W e propose a new model for supervised learning to rank. In our model, the relevancy labels are are assumed to follow a categorical distribution whose probabilities are constru cted based on a scoring function. Learning - to-rank methods can generally be categorized into pointwis e, pairwise, and listwise approaches. Our approach belongs to the class of pointwise methods. Although it has previously been reported that pointwise methods cannot achieve as good performance as of pairwise or listwise approaches, we show that the proposed method achieves better or comparable results on two datasets compared with pairwise and listwise methods. Index T erms-- Learning to rank, Monte Carlo Gradient Estimation, Deep learning 1. INTRODUCTION Learning to rank is fundamental to information retrieval, E-commerce, and many other applications, for ranking items [1].