A Dataset of Peer Reviews (PeerRead): Collection, Insights and NLP Applications
This dataset enables research into peer review processes, benefiting the scientific community, but it is incremental as it primarily provides a new resource rather than a breakthrough method.
The authors introduced PeerRead v1, the first public dataset of 14.7K paper drafts and 10.7K peer reviews from top-tier venues, and demonstrated that simple models can predict paper acceptance with up to 21% error reduction and outperform baselines in scoring review aspects like originality and impact.
Peer reviewing is a central component in the scientific publishing process. We present the first public dataset of scientific peer reviews available for research purposes (PeerRead v1) providing an opportunity to study this important artifact. The dataset consists of 14.7K paper drafts and the corresponding accept/reject decisions in top-tier venues including ACL, NIPS and ICLR. The dataset also includes 10.7K textual peer reviews written by experts for a subset of the papers. We describe the data collection process and report interesting observed phenomena in the peer reviews. We also propose two novel NLP tasks based on this dataset and provide simple baseline models. In the first task, we show that simple models can predict whether a paper is accepted with up to 21% error reduction compared to the majority baseline. In the second task, we predict the numerical scores of review aspects and show that simple models can outperform the mean baseline for aspects with high variance such as 'originality' and 'impact'.