FairJudge: An Adaptive, Debiased, and Consistent LLM-as-a-Judge

TL;DR

FairJudge is a novel LLM-based evaluation system that adaptively learns to provide unbiased, consistent judgments across different evaluation modes, significantly improving agreement and reducing biases.

Contribution

It introduces a learnable, regularized judging policy and a high-information-density dataset, enabling adaptive, debiased, and consistent evaluations in LLM-as-a-Judge systems.

Findings

Improves agreement and F1 scores on benchmarks

Reduces non-semantic biases in judgments

Outperforms larger instruction-tuned LLMs

Abstract

Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to task- and domain-specific evaluation criteria, systematic biases driven by non-semantic cues such as position, length, format, and model provenance, and evaluation inconsistency that leads to contradictory judgments across different evaluation modes (e.g., pointwise versus pairwise). To address these issues, we propose FairJudge, an adaptive, debiased, and consistent LLM-as-a-Judge. Unlike prior approaches that treat the judge as a static evaluator, FairJudge models judging behavior itself as a learnable and regularized policy. From a data-centric perspective, we construct a high-information-density judging dataset that explicitly injects supervision signals aligned with evaluation behavior. Building on this dataset, we adopt a curriculum-style SFT-DPO-GRPO training paradigm that progressively aligns rubric adherence, bias mitigation, and cross-mode consistency, while avoiding catastrophic forgetting. Experimental results on multiple internal and public benchmarks show that FairJudge consistently improves agreement and F1, reduces non-semantic biases, and outperforms substantially larger instruction-tuned LLMs. All resources will be publicly released after acceptance to facilitate future research.