Debiasing Neural Retrieval via In-batch Balancing Regularization

TL;DR

This paper introduces In-Batch Balancing Regularization (IBBR), a novel method that uses a differentiable fairness metric based on T-statistics to reduce bias in neural retrieval models while maintaining high ranking accuracy.

Contribution

The paper proposes a new regularization technique, IBBR, with a differentiable fairness measure (nPRF) that directly optimizes for reduced bias in neural IR models.

Findings

IBBR significantly reduces bias in neural rankers.

Minimal impact on ranking performance compared to baseline.

Effective on MS MARCO Passage Retrieval dataset.

Abstract

People frequently interact with information retrieval (IR) systems, however, IR models exhibit biases and discrimination towards various demographics. The in-processing fair ranking methods provide a trade-offs between accuracy and fairness through adding a fairness-related regularization term in the loss function. However, there haven't been intuitive objective functions that depend on the click probability and user engagement to directly optimize towards this. In this work, we propose the In-Batch Balancing Regularization (IBBR) to mitigate the ranking disparity among subgroups. In particular, we develop a differentiable \textit{normed Pairwise Ranking Fairness} (nPRF) and leverage the T-statistics on top of nPRF over subgroups as a regularization to improve fairness. Empirical results with the BERT-based neural rankers on the MS MARCO Passage Retrieval dataset with the human-annotated non-gendered queries benchmark \citep{rekabsaz2020neural} show that our IBBR method with nPRF achieves significantly less bias with minimal degradation in ranking performance compared with the baseline.