ReviewMT

该数据集是一个全面的数据集，包含了超过26,841篇论文及92,017条评审意见，旨在助力大型语言模型在同行评审过程中的多轮对话应用。该数据集被划分为两个子集：ReviewMT-ICLR和ReviewMT-NC，分别体现了ICLR和《自然通讯》评审流程的独特特点。规模上，该数据集涵盖了超过26,841篇论文和92,017条评审意见，其任务重点在于模拟同行评审中的多轮对话。This is a comprehensive dataset containing over 26,841 papers and 92,017 review comments, aiming to facilitate the application of large language models (LLMs) in multi-turn dialogue scenarios during the peer review process. The dataset is divided into two subsets: ReviewMT-ICLR and ReviewMT-NC, which respectively reflect the unique characteristics of the peer review workflows of ICLR and *Nature Communications*. In terms of scale, this dataset covers over 26,841 papers and 92,017 review comments, with its task focus on simulating multi-turn dialogues in peer review processes.

搜集汇总

背景与挑战

背景概述

ReviewMT是一个学术同行评审数据集，将评审过程重新定义为多轮、长上下文对话，包含作者、审稿人和决策者等角色。该数据集涵盖超过26,841篇论文和92,017条评审，来自ICLR和Nature Communications等顶级会议和期刊，旨在支持大语言模型进行动态对话模拟和评估。