Mitigating Selection Bias in Large Language Models via Permutation-Aware GRPO

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zheng, Jinquan, Yuan, Jia, Yao, Jiacheng, Gu, Chenyang, Zheng, Pujun, He, Guoxiu
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914519367811072
author Zheng, Jinquan
Yuan, Jia
Yao, Jiacheng
Gu, Chenyang
Zheng, Pujun
He, Guoxiu
author_facet Zheng, Jinquan
Yuan, Jia
Yao, Jiacheng
Gu, Chenyang
Zheng, Pujun
He, Guoxiu
contents Large language models (LLMs) used for multiple-choice and pairwise evaluation tasks often exhibit selection bias due to non-semantic factors like option positions and label symbols. Existing inference-time debiasing is costly and may harm reasoning, while pointwise training ignores that the same question should yield consistent answers across permutations. To address this issue, we propose Permutation-Aware Group Relative Policy Optimization (PA-GRPO), which mitigates selection bias by enforcing permutation-consistent semantic reasoning. PA-GRPO constructs a permutation group for each instance by generating multiple candidate permutations, and optimizes the model using two complementary mechanisms: (1) cross-permutation advantage, which computes advantages relative to the mean reward over all permutations of the same instance, and (2) consistency-aware reward, which encourages the model to produce consistent decisions across different permutations. Experimental results demonstrate that PA-GRPO outperforms strong baselines across seven benchmarks, substantially reducing selection bias while maintaining high overall performance. The code is available on github (https://github.com/ECNU-Text-Computing/PA-GRPO).
format Preprint
id arxiv_https___arxiv_org_abs_2603_21016
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Mitigating Selection Bias in Large Language Models via Permutation-Aware GRPO
Zheng, Jinquan
Yuan, Jia
Yao, Jiacheng
Gu, Chenyang
Zheng, Pujun
He, Guoxiu
Computation and Language
Artificial Intelligence
Machine Learning
Large language models (LLMs) used for multiple-choice and pairwise evaluation tasks often exhibit selection bias due to non-semantic factors like option positions and label symbols. Existing inference-time debiasing is costly and may harm reasoning, while pointwise training ignores that the same question should yield consistent answers across permutations. To address this issue, we propose Permutation-Aware Group Relative Policy Optimization (PA-GRPO), which mitigates selection bias by enforcing permutation-consistent semantic reasoning. PA-GRPO constructs a permutation group for each instance by generating multiple candidate permutations, and optimizes the model using two complementary mechanisms: (1) cross-permutation advantage, which computes advantages relative to the mean reward over all permutations of the same instance, and (2) consistency-aware reward, which encourages the model to produce consistent decisions across different permutations. Experimental results demonstrate that PA-GRPO outperforms strong baselines across seven benchmarks, substantially reducing selection bias while maintaining high overall performance. The code is available on github (https://github.com/ECNU-Text-Computing/PA-GRPO).
title Mitigating Selection Bias in Large Language Models via Permutation-Aware GRPO
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2603.21016