Blind to the Human Touch: Overlap Bias in LLM-Based Summary Evaluation

TL;DR

This paper investigates biases in LLM-based summary evaluation, revealing that LLM judges favor LLM-generated summaries over human ones as overlap decreases, highlighting limitations in current evaluation methods.

Contribution

The study provides a detailed bias analysis of LLM judges based on overlap metrics, across multiple models and in the context of summarization evaluation.

Findings

LLM judges prefer LLM summaries over human summaries as overlap decreases.

Models struggle to evaluate summaries with limited overlap, indicating a need for improved techniques.

Bias patterns are consistent across different LLMs regardless of their own biases.

Abstract

Large language model (LLM) judges have often been used alongside traditional, algorithm-based metrics for tasks like summarization because they better capture semantic information, are better at reasoning, and are more robust to paraphrasing. However, LLM judges show biases for length and order among others, and are vulnerable to various adversarial input prompts. While recent studies have looked into these biases, few have analyzed them at a more granular level in relation to a well-defined overlap metric. In this work we provide an LLM judge bias analysis as a function of overlap with human-written responses in the domain of summarization. We test 9 recent LLMs with parameter counts ranging from 1 billion to 12 billion, including variants of Gemma 3 and LLaMA 3. We find that LLM judges increasingly prefer summaries generated by other LLMs over those written by humans as the similarities (as measured by ROUGE and BLEU) between the judged summaries decrease, and this pattern extends to all but one model tested, and exists regardless of the models' own position biases. Additionally, we find that models struggle to judge even summaries with limited overlaps, suggesting that LLM-as-a-judge in the summary domain should rely on techniques beyond a simple comparison.