AI-written peer reviews dataset
收藏资源简介:
本研究引入了一个包含788,984个由五种广泛使用的大型语言模型生成的AI撰写同行评审以及相应的人类撰写的同行评审的数据集。该数据集是迄今为止评估AI文本在同行评审中检测性的最大资源,覆盖了两大会AI研究会议(ICLR和NeurIPS)八年来的论文。数据集通过整合现有来源的人类撰写的同行评审和针对同一篇论文使用五种最先进的LLM生成的AI撰写的同行评审而构建。
This study introduces a dataset consisting of 788,984 pairs of AI-generated peer reviews (produced by five widely used large language models) and their corresponding human-written peer reviews. As the largest available resource to date for evaluating the detectability of AI-generated text in peer reviews, this dataset covers papers from eight years of two top-tier AI research conferences: ICLR and NeurIPS. The dataset was constructed by integrating human-written peer reviews from existing sources and AI-generated peer reviews for the same papers using five state-of-the-art LLMs.

- 1Is Your Paper Being Reviewed by an LLM? A New Benchmark Dataset and Approach for Detecting AI Text in Peer ReviewIntel Labs · 2025年



