Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Peer Review as A Multi-Turn and Long-Context Dialogue with Role-Based Interactions

Cheng Tan, Dongxin Lyu, Siyuan Li, Zhangyang Gao, Jingxuan Wei, Siqi Ma, Zicheng Liu, Stan Z. Li

TL;DR

This work reframes academic peer review as a long-context, multi-turn dialogue among authors, reviewers, and decision makers, and introduces ReviewMT—a dataset built from ICLR and Nature Communications—to simulate complete, iterative peer-review interactions. It formalizes the triadic workflow with explicit turns: initial reviews, author rebuttals, final reviews, and meta decisions, and pioneers a suite of evaluation metrics tailored to each role. Through experiments on multiple open-source LLMs, the study shows that supervised fine-tuning significantly improves performance across validity and quality metrics, highlighting the value of role-based dialogue for scalable, fair peer review. The dataset and framework offer a robust foundation for future research on LLM-assisted peer review and open avenues for improving efficiency and transparency in scholarly publishing.

Abstract

Large Language Models (LLMs) have demonstrated wide-ranging applications across various fields and have shown significant potential in the academic peer-review process. However, existing applications are primarily limited to static review generation based on submitted papers, which fail to capture the dynamic and iterative nature of real-world peer reviews. In this paper, we reformulate the peer-review process as a multi-turn, long-context dialogue, incorporating distinct roles for authors, reviewers, and decision makers. We construct a comprehensive dataset containing over 26,841 papers with 92,017 reviews collected from multiple sources, including the top-tier conference and prestigious journal. This dataset is meticulously designed to facilitate the applications of LLMs for multi-turn dialogues, effectively simulating the complete peer-review process. Furthermore, we propose a series of metrics to evaluate the performance of LLMs for each role under this reformulated peer-review setting, ensuring fair and comprehensive evaluations. We believe this work provides a promising perspective on enhancing the LLM-driven peer-review process by incorporating dynamic, role-based interactions. It aligns closely with the iterative and interactive nature of real-world academic peer review, offering a robust foundation for future research and development in this area. We open-source the dataset at https://github.com/chengtan9907/ReviewMT.

Peer Review as A Multi-Turn and Long-Context Dialogue with Role-Based Interactions

TL;DR

This work reframes academic peer review as a long-context, multi-turn dialogue among authors, reviewers, and decision makers, and introduces ReviewMT—a dataset built from ICLR and Nature Communications—to simulate complete, iterative peer-review interactions. It formalizes the triadic workflow with explicit turns: initial reviews, author rebuttals, final reviews, and meta decisions, and pioneers a suite of evaluation metrics tailored to each role. Through experiments on multiple open-source LLMs, the study shows that supervised fine-tuning significantly improves performance across validity and quality metrics, highlighting the value of role-based dialogue for scalable, fair peer review. The dataset and framework offer a robust foundation for future research on LLM-assisted peer review and open avenues for improving efficiency and transparency in scholarly publishing.

Abstract

Large Language Models (LLMs) have demonstrated wide-ranging applications across various fields and have shown significant potential in the academic peer-review process. However, existing applications are primarily limited to static review generation based on submitted papers, which fail to capture the dynamic and iterative nature of real-world peer reviews. In this paper, we reformulate the peer-review process as a multi-turn, long-context dialogue, incorporating distinct roles for authors, reviewers, and decision makers. We construct a comprehensive dataset containing over 26,841 papers with 92,017 reviews collected from multiple sources, including the top-tier conference and prestigious journal. This dataset is meticulously designed to facilitate the applications of LLMs for multi-turn dialogues, effectively simulating the complete peer-review process. Furthermore, we propose a series of metrics to evaluate the performance of LLMs for each role under this reformulated peer-review setting, ensuring fair and comprehensive evaluations. We believe this work provides a promising perspective on enhancing the LLM-driven peer-review process by incorporating dynamic, role-based interactions. It aligns closely with the iterative and interactive nature of real-world academic peer review, offering a robust foundation for future research and development in this area. We open-source the dataset at https://github.com/chengtan9907/ReviewMT.
Paper Structure (12 sections, 2 equations, 6 figures, 3 tables)

This paper contains 12 sections, 2 equations, 6 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Instruction Tuning Dataset
  4. LLM in Review
  5. Preliminaries
  6. ReviewMT Dataset
  7. Data Source
  8. Data Processing
  9. Dataset Statistics
  10. Evaluation Metrics
  11. Experiments
  12. Conclusion and Limitation

Figures (6)

  • Figure 1: Comparison of existing LLM applications in peer review and our reformulated framework.
  • Figure 2: Overview of the data processing pipeline for the ReviewMT dataset.
  • Figure 3: Statistics of the ICLR papers and reviews in the ReviewMT-ICLR dataset.
  • Figure 4: The word cloud of the keywords in the ReviewMT dataset.
  • Figure 5: The radar chart of text similarity metrics for LLMs on the ReviewMT-ICLR dataset.
  • ...and 1 more figures