PeerQA-XT
收藏资源简介:
PeerQA-XT 是一个包含 12,628 个自由形式科学问答对的数据集,源自十个科学领域的学术文章及其同行评审。该数据集受 PeerQA 启发,但通过更广泛和全面的科学文章和同行评审集合,以及使用合成管道进行问题和答案提取,进一步扩展了原始数据集。问题由大型语言模型从同行评审中自动提取,对应的答案则来自作者的反驳或直接从论文文本生成。数据集包含以下字段:`pid`(论文唯一标识符)、`qid`(问题唯一标识符)、`question`(从同行评审中提取的问题)、`answer`(对应的答案)、`paper`(科学论文全文)、`domain`(论文所属科学领域)。数据集分为训练集(10,128 个样本)、验证集(1,248 个样本)和测试集(1,252 个样本)。PeerQA-XT 适用于问答和文本生成任务,旨在克服以往科学问答数据集在规模和质量上的局限性,通过利用同行评审中的专家级问题和作者反驳中的精确答案,结合自动验证流程,实现高质量和大规模的数据集构建。
PeerQA-XT is a dataset containing 12,628 free-form scientific question-answer pairs sourced from academic articles and their peer reviews across ten scientific disciplines. Inspired by the original PeerQA dataset, PeerQA-XT further expands the initial dataset through a broader and more comprehensive collection of scientific articles and peer reviews, alongside a synthetic pipeline for question and answer extraction. Questions are automatically extracted from peer reviews by large language models, while corresponding answers are derived from authors' rebuttals or directly generated from the full text of the papers. The dataset includes the following fields: `pid` (unique paper identifier), `qid` (unique question identifier), `question` (question extracted from peer reviews), `answer` (corresponding answer), `paper` (full text of the scientific paper), and `domain` (scientific discipline of the paper). The dataset is split into a training set (10,128 samples), a validation set (1,248 samples), and a test set (1,252 samples). PeerQA-XT is applicable to question answering and text generation tasks, aiming to overcome the limitations in scale and quality of previous scientific question answering datasets. By leveraging expert-level questions from peer reviews and precise answers from authors' rebuttals combined with an automatic verification process, it enables the construction of high-quality and large-scale datasets.



