CatRAG: Functor-Guided Structural Debiasing with Retrieval Augmentation for Fair LLMs

Ravi Ranjan; Utkarsh Grover; Mayur Akewar; Xiaomin Lin; Agoritsa Polyzou

cs.CLMar 23, 2026

CatRAG: Functor-Guided Structural Debiasing with Retrieval Augmentation for Fair LLMs

Authors: Ravi Ranjan, Utkarsh Grover, Mayur Akewar, Xiaomin Lin, Agoritsa Polyzou

Organizations: Florida International University Miami, USA · University of South Florida Tampa, USA

Abstract

Large Language Models (LLMs) are deployed in high-stakes settings but can show demographic, gender, and geographic biases that undermine fairness and trust. Prior debiasing methods, including embedding-space projections, prompt-based steering, and causal interventions, often act at a single stage of the pipeline, resulting in incomplete mitigation and brittle utility trade-offs under distribution shifts. We propose CatRAG Debiasing, a dual-pronged framework that integrates functor with Retrieval-Augmented Generation (RAG) guided structural debiasing. The functor component leverages category-theoretic structure to induce a principled, structure-preserving projection that suppresses bias-associated directions in the embedding space while retaining task-relevant semantics. On the Bias Benchmark for Question Answering (BBQ) across three open-source LLMs (Meta Llama-3, OpenAI GPT-OSS, and Google Gemma-3), CatRAG achieves state-of-the-art results, improving accuracy by up to 40% over the corresponding base models and by more than 10% over prior debiasing methods, while reducing bias scores to near zero (from 60% for the base models) across gender, nationality, race, and intersectional subgroups.

Figures & tables

Fig. 1: Research challenge: the same job-advice query yields systematically different recommendations when the context implies developed vs. developing countries, reflecting stereotyped associations rather than qualification-based reasoning. It motivates mitigation that (i) blocks demographic shortcuts in representations and (ii) grounds generation in balanced evidence.

Fig. 2: Overview of the proposed pipeline. The input query is processed along two paths: (1) Functor-guided structural debiasing maps the biased embedding space to an unbiased one via a debiased projection, reducing demographic separability while preserving task-relevant structure; (2) Retrieval augmentation selects a small set of diverse, counter-stereotypical evidence passages from an external corpus. A context fusion module injects retrieved evidence into the prompt, and the LLM generates using the projected embedding layer to produce a grounded, fair output.

Fig. 3: Gender subset scatter plots: x-axis is the confidence score for the male-coded option and y-axis for the female-coded option. Red points are the base model; colored points show the post-mitigation distribution for each method.

Method	Acc.	AccImpr	BS	BS-Impr
Base Model	48.9%	0.0	0.63	0.0
CE Debiasing [ 18 ]	64.2%	+15.3%	0.28	+54.3%
Self Debiasing [ 16 ]	59.8%	+10.9%	0.41	+35.1%
SP Debiasing [ 15 ]	68.5%	+19.6%	0.19	+69.2%
Causal Debiasing [ 17 ]	78.6%	+28.2%	0.10	+84.1%
CatRAG (Ours)	80.7%	+32.3%	0.01	+97.6%

TABLE I: Overall performance on BBQ (4,000 questions) across all subsets on Llama-3 model. Accuracy Improvement (AccImpr) is relative to base model accuracy (48.9%). Bias Score(BS) ranges from

-1

+1

with

0

ideal.

		Llama-3		GPT-OSS		Gemma-3
Category	Method	Acc.	BS	Acc.	BS	Acc.	BS
Gender	Base	52.3	0.59	56.8	0.52	53.6	0.57
	CE [ 18 ]	67.8	0.25	72.0	0.20	69.1	0.23
	Self [ 16 ]	62.4	0.38	66.9	0.32	63.7	0.36
	SP [ 15 ]	71.2	0.16	75.6	0.12	72.6	0.14
	Causal [ 17 ]	78.1	0.10	82.4	0.04	76.4	0.09

TABLE II: Accuracy (Acc.) and Bias Score (BS) results on BBQ subset (1k Qs/category) for Llama-3-1B , GPT-OSS-20B , and Gemma-3-1B-IT .

Fig. 4: Accuracy vs. Bias Score for Functor-only, RAG-only, and full pipeline. Better performance lies toward the upper-left (higher accuracy, lower bias score).

Variant	Acc. (%)	BS
Base	$48.9\pm 1.5$	$0.63\pm 0.03$
Functor-only	$70.5\pm 1.2$	$0.15\pm 0.02$
RAG-only	$65.3\pm 1.4$	$0.24\pm 0.03$
Full CatRAG	$\mathbf{81.2\pm 1.0}$	$\mathbf{0.01\pm 0.01}$
$d_{u}{=}128$	$78.9\pm 1.2$	$0.02\pm 0.01$
$d_{u}{=}512$	$80.7\pm 1.1$	$0.01\pm 0.01$

TABLE III: Ablation & sensitivity on BBQ Gender (1,000 qs) with Llama-3. Accuracy (Acc.) and Bias Score (BS) Mean

\pm

95% CI over 3 seeds.

Language Model Debiasing

See all 26 papers

DebiasRAG: A Tuning-Free Path to Fair Generation in Large Language Models through Retrieval-Augmented Generation
May 15, 2026Rui Chu, Bingyin Zhao, Thanh Quoc Hung Le +6Retrieval-Augmented Generation Debiased ML
Position: LLMs Must Use Functor-Based and RAG-Driven Bias Mitigation for Fairness
Mar 7, 2026Ravi Ranjan, Utkarsh Grover, Agorista PolyzouSocial Bias in Language Models Demographic Bias in Language Models
Knight Foundation School of Computing and Information Sciences, Florida International University, Miami, FL 33199 · College of Engineering, University of South Florida, Tampa, FL 33620
Knowing Bias, Doing Better: Mitigating Social Bias in LLMs via Know-Bias Neuron Enhancement
Jan 29, 2026Jinhao Pan, Chahat Raj, Anjishnu Mukherjee +4Debiased ML Social Bias in Language Models
Department of Computer Science, George Mason University, Fairfax, VA, USA · Lightridge High School, Aldie, VA, USA

May 15, 2026J/K move · Enter open · S save

DebiasRAG: A Tuning-Free Path to Fair Generation in Large Language Models through Retrieval-Augmented Generation

Rui Chu, Bingyin Zhao, Thanh Quoc Hung Le, Duy Cao Hoang+5

Retrieval-Augmented Generation Debiased ML Social Bias in Language Models Language Model Debiasing

Open paper AskPDF arXiv

Large language models (LLMs) have achieved unprecedented success due to their exceptional generative capabilities. However, because they depend on knowledge encapsulated from training corpora, they may produce hallucinations, stereotypes, and socially biased content. In particular, LLMs are prone to prejudiced responses involving race, gender, and age, which are collectively referred to as social biases. Prior studies have used fine-tuning and prompt engineering to mitigate such biases in LLMs, but these methods require additional training resources or domain knowledge to design the framework. Moreover, they may degrade the original capabilities of LLMs and often overlook the need for dynamic debiasing contexts for fairer inference. In this paper, we propose DebiasRAG, a novel tuning-free and dynamic query-specific debiasing framework based on retrieval-augmented generation (RAG). DebiasRAG improves fairness while preserving the intrinsic properties of LLMs, such as representation ability. DebiasRAG consists of three stages: (1) query-specific debiasing candidate generation; (2) context candidate pool construction; and (3) gradient-updated debiasing-guided context piece reranking. First, DebiasRAG leverages self-diagnosed bias contexts relevant to the query through regular retrieval, where the bias contexts are prepared offline by the DebiasRAG provider. Given the query-specific bias contexts, DebiasRAG reversely produces debiasing contexts, which are provided as additional fairness constraints for LLM outputs. Second, a regular RAG retrieval process produces query-related contexts from the regular RAG document database, such as a chunked Wikipedia dataset.

Abstract

Language Model Debiasing

Explore similar work

DebiasRAG: A Tuning-Free Path to Fair Generation in Large Language Models through Retrieval-Augmented Generation

Position: LLMs Must Use Functor-Based and RAG-Driven Bias Mitigation for Fairness

Knowing Bias, Doing Better: Mitigating Social Bias in LLMs via Know-Bias Neuron Enhancement