PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers

2026-05-26

AI Summary

This study addresses the lack of systematic evaluation of large language models (LLMs) in scientific peer review, particularly regarding their ability to identify scientific flaws and conduct in-depth analysis. The authors propose PRISM, a novel benchmark framework that integrates argument mining, retrieval-augmented verification, and consensus-based scoring to comprehensively assess LLMs and human reviewers across four dimensions: analytical depth, novelty evaluation, defect identification and prioritization, and multidimensional constructiveness. Results show that while LLMs can match or even surpass human performance on individual tasks, no existing system yet matches human reviewers across all dimensions simultaneously. These findings highlight the potential of LLMs as specialized assistive tools in peer review rather than as full replacements for human judgment.

Abstract

The rapid growth in submissions to machine learning venues has strained the scientific peer-review system and intensified interest in LLM-based automated peer reviewers. However, how good these systems are actually, especially compared to human reviewers at catching scientific gaps, remains poorly understood. In this work, we introduce PRISM (Peer Review Intelligence via Structured Multi-dimensional assessment), a benchmarking framework that evaluates review quality across four dimensions: Depth of Analysis, Novelty Assessment,Flaw Identification & Major Issues Prioritization, and Multi-dimensional Constructiveness. Unlike most existing evaluations based on surface-level metrics like ROUGE and BLEU, or unconstrained LLM-as-a-judge prompting that conflates fluency with rigor, PRISM grounds each dimension in argument mining, retrieval-augmented verification, and consensus-based scoring. We apply PRISM to benchmark five leading automated reviewer systems and human reviewers on a stratified corpus of reviews from ICLR, ICML, and NeurIPS. The results reveal that LLMs can match or beat human reviewers on individual dimensions: comparable depth of analysis, stronger novelty verification, and highly accurate critique prioritization. However, no single system consistently matches the balanced performance of the human baseline across all dimensions at once. Each exhibits a distinct specialization profile with characteristic blind spots -- failure modes that aggregate metrics miss entirely. The implication is that LLM reviewers are best understood as targeted supplements to human review, effective within specific dimensions, but unreliable as standalone replacements. Our demo and key results can be found at https://khanhthanhdev.github.io/prism-page/.

Problem

• LLM peer review
• review quality evaluation
• scientific gaps
• benchmarking
• peer review dimensions

Innovation

• PRISM
• argument mining
• retrieval-augmented verification
• multi-dimensional evaluation
• LLM peer review