UKPLab/ProReviewer-Dataset
收藏资源简介:
该数据集是一个版本匹配的数据集,包含ICLR论文提交与其初始同行评审和评分。每个样本包括论文的初始提交内容,以及评审者在讨论阶段之前给出的初始评分,支持自动化学术同行评审的研究。数据集用于训练和评估自动同行评审系统,所有论文和评审都是版本匹配的:论文内容对应初始提交,评审分数反映初始评分,确保模型学习评估首次提交的论文,评分反映第一印象而非讨论后的调整。数据集分为训练集(来自ICLR 2025,4,011个样本)和测试集(来自ICLR 2026,1,000个样本),特征包括论文ID、arXiv ID、标题、Markdown内容、评审列表、聚合评分、元数据、元评审和最终决定等。
A version-matched dataset of ICLR paper submissions paired with their initial peer reviews and scores. Each sample contains the initial submission of a paper matched with the reviewers initial ratings (before the discussion phase), enabling research on automated scientific peer review. This dataset supports training and evaluating automated peer review systems. All papers and reviews are version-matched: the paper content corresponds to the initial submission, and the review scores reflect the initial ratings assigned before any author-reviewer discussion. This version matching is important for training review agents, as it ensures the model learns to evaluate papers as they were first submitted, with scores that reflect first impressions rather than post-discussion adjustments.




