PeerSum

该数据集名为PeerSum，它包含了从OpenReview.net抓取的两个计算机科学国际会议（ICLR和NeurIPS）的同行评审数据。这些数据包括论文摘要、官方评审和公开讨论，最终形成了作为总结的元评审。PeerSum数据集还特别考虑了文档之间的层次关系，并包含了诸如论文接受情况和评审评分等元数据。该数据集旨在促进对元评审生成的研究，并解决评审中存在的信息冲突问题。其规模包含14,993个样本（训练集/验证集/测试集分别为11,995/1,499/1,499个样本），任务聚焦于元评审的生成和总结。

The dataset is named PeerSum. It consists of peer review data scraped from OpenReview.net for two premier computer science international conferences, ICLR and NeurIPS. The data includes paper abstracts, official reviews, and public discussions, which ultimately constitute the meta-reviews serving as the official summaries of the submissions. Additionally, the PeerSum dataset specifically accounts for the hierarchical relationships among different documents, and incorporates metadata such as paper acceptance status and review scores. This dataset is designed to facilitate research on meta-review generation and resolve the problem of information conflicts within peer reviews. It comprises a total of 14,993 samples, with the training, validation, and test splits containing 11,995, 1,499, and 1,499 samples respectively. The core tasks of this dataset focus on meta-review generation and summarization.

搜集汇总

背景与挑战

背景概述

PeerSum是一个多文档摘要数据集，专注于生成科学论文的元评论，包含来自OpenReview的同行评审数据。其特点包括层次化的对话结构、丰富的元数据（如评审评分和论文接受结果），以及处理源文档中可能存在的冲突信息的能力。