ARSM Gradient Estimator for Supervised Learning to Rank

TL;DR

This paper introduces a novel gradient estimator for supervised learning to rank that handles non-differentiable loss functions and improves performance on benchmark datasets.

Contribution

It presents a new model with an unbiased, low-variance gradient estimator for categorical relevance labels, enhancing flexibility in loss function choice.

Findings

Achieves better or comparable results on two datasets.

Supports non-differentiable loss functions.

Flexible framework for pointwise, pairwise, and listwise learning-to-rank methods.

Abstract

We propose a new model for supervised learning to rank. In our model, the relevance labels are assumed to follow a categorical distribution whose probabilities are constructed based on a scoring function. We optimize the training objective with respect to the multivariate categorical variables with an unbiased and low-variance gradient estimator. Learning-to-rank methods can generally be categorized into pointwise, pairwise, and listwise approaches. Although our scoring function is pointwise, the proposed framework permits flexibility over the choice of the loss function. In our new model, the loss function need not be differentiable and can either be pointwise or listwise. Our proposed method achieves better or comparable results on two datasets compared with existing pairwise and listwise methods.