ARIES

ARIES数据集由Allen人工智能研究所创建，包含196条人工标注的科学论文评审意见及其相应的编辑内容。数据集大小为3.9K条评审意见，通过自动匹配编辑内容获得。创建过程涉及从OpenReview获取论文、评审和作者回应，并手动识别可操作的评审意见及其对应的编辑。ARIES数据集主要用于训练和评估大型语言模型在科学论文修订任务中的表现，旨在解决模型在理解评审意见并生成相应编辑方面的挑战。The ARIES dataset was developed by the Allen Institute for Artificial Intelligence. It encompasses 196 manually annotated scientific paper review comments and their corresponding editorial revisions. The dataset comprises 3.9K review comments acquired through automatic matching with editorial content. The construction of the dataset involves retrieving papers, reviews, and author responses from OpenReview, followed by manual identification of actionable review comments and their matched editorial revisions. The ARIES dataset is primarily utilized for training and evaluating the performance of large language models (LLMs) on scientific paper revision tasks, with the goal of addressing the challenges that models encounter when understanding review comments and generating corresponding editorial revisions.

搜集汇总

背景与挑战

背景概述

ARIES是一个科学论文编辑数据集，包含根据同行评审意见进行的修改记录，用于研究编辑与评审之间的对齐关系。数据集提供论文编辑、评审评论和对齐标签，支持模型训练和推理，以分析科学写作中的修订过程。