Learning an Efficient Multi-Turn Dialogue Evaluator from Multiple LLM Judges

TL;DR

This paper introduces a computationally efficient multi-turn dialogue evaluator that consolidates multiple LLM judges' preferences into a single model, improving evaluation accuracy and speed.

Contribution

The authors propose a novel method to aggregate multiple LLM judges' preferences into one model, reducing inference costs while maintaining evaluation quality.

Findings

Outperforms existing baselines on seven dialogue evaluation benchmarks.

Achieves faster evaluation with comparable or better accuracy.

Demonstrates robustness across diverse dialogue scenarios.

Abstract

Evaluating the conversational abilities of large language models (LLMs) remains a challenging task. Current mainstream approaches primarily rely on the "LLM-as-a-judge" paradigm, where an LLM is prompted to serve as an evaluator to assess dialogue quality. However, such methods often suffer from various biases, which undermine the reliability and consistency of the evaluation results. To mitigate these biases, recent methods employ multiple LLMs as judges and aggregate their judgments to select the optimal assessment. Although effective, this multi-judge approach incurs significant computational overhead during inference. In this paper, we propose an efficient dialogue evaluator that captures the collective wisdom of multiple LLM judges by aggregating their preference knowledge into a single model. Our approach preserves the advantages of diverse multi-judge feedback while drastically reducing the evaluation cost, enabling fast, flexible, and fine-grained dialogue quality assessment. Extensive experiments on seven single rating and pairwise comparison dialogue evaluation benchmarks demonstrate that our method outperforms existing baselines across diverse scenarios, showcasing its efficiency and robustness.