Lune

ACL Findings2026

Fair RAG: End-to-End Fairness Across Retrieval and Generation

Farsheed Haque, Zhe Fu, Ramit Aditya, Depeng Xu, Xi Niu

Abstract

Large Language Models (LLMs) used in Retrieval-Augmented Generation (RAG) can amplify demographic bias: retrievers may surface skewed context and generators can propagate that skew into decisions. Prior work typically treats fairness in retrieval or generation in isolation, leaving end-to-end fairness in RAG underexplored. We propose a post-hoc pipeline that jointly controls both stages: (i) a Fair Greedy Reranker (FGR) that builds prefix-balanced slates toward a target group mix; (ii) a Residual Slate Bias Estimator (RSBE) using signed, prefix-sensitive normalized discounted cumulative KL divergence (NDKL) to quantify remaining skew; and (iii) Confidence-Gated Logit Calibration (CGLC) that converts the residual signal into small and margin-focused logit corrections without retraining. On an occupation classification task, our approach reduces retriever-side skew (lowest NDKL among baselines for both dense and sparse retrievers) and achieves the lowest generator-side disparity (e.g., Risk Difference) while largely preserving utility. The same calibration can be tuned to alternative fairness criteria (e.g., Equal Opportunity) with minimal utility loss.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • No Free Lunch: Retrieval-Augmented Generation Undermines Fairness in LLMs, Even for Vigilant Users cites Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
  • Mitigating Bias in RAG: Controlling the Embedder cites No Free Lunch: Retrieval-Augmented Generation Undermines Fairness in LLMs, Even for Vigilant Users
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on3

  1. Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksPatrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni et al.NeurIPS 2020 · 19,162 citations
  2. No Free Lunch: Retrieval-Augmented Generation Undermines Fairness in LLMs, Even for Vigilant UsersMengxuan Hu, Hongyi Wu, Ronghang Zhu, Zihan Guan et al.EMNLP Findings 2025 · 9 citations
  3. Mitigating Bias in RAG: Controlling the EmbedderTaeyoun Kim, Jacob Mitchell Springer, Aditi Raghunathan, Maarten SapACL Findings 2025 · 5 citations

Related papers

  1. Reliable Decision‑Making via Calibration‑Oriented Retrieval‑Augmented GenerationChaeyun Jang, Deukhwan Cho, Seanie Lee, Hyungi Lee et al.NeurIPS 2025 · 5 citations
  2. FairRAG: Fair Human Generation via Fair Retrieval AugmentationRobik Shrestha, Yang Zou, Qiuyu Chen, Zhiheng Li et al.CVPR 2024 · 6 citations
  3. FAIR-RAG: An End-to-End Framework for Mitigating Political Bias through Fair Retrieval-Augmented GenerationJaebeom You, Kisung Lee, Hyuk-Yoon KwonSIGIR 2026
  4. Omnia: Efficient RAG Serving through Speculative SchedulingRongtian Fu, Shigang Li, Youxuan Xu, Tong Wu et al.HPDC 2026
  5. On Optimal Steering to Achieve Exact FairnessMohit Sharma, Amit Deshpande, Chiranjib Bhattacharyya, Rajiv Ratn ShahNeurIPS 2025 · 2 citations
Dusk over the sea between two cliffs drawn in fine vertical lines