README.md

metadata

license: apache-2.0
task_categories:
  - text-generation
language:
  - en
size_categories:
  - 10K<n<100K

Raw Review Dataset for Reviewer2

This is the raw version of our dataset. The cleaned data files that can be directly used for fine-tuning is in this directory.

Dataset Structure

The folders are structured in the following way:

venue
|--venue_year
   |--venue_year_metadata
      |--venue_year_id1_metadata.json
      |--venue_year_id2_metadata.json
      ...
   |--venue_year_paper
      |--venue_year_id1_paper.json
      |--venue_year_id2_paper.json
      ...
   |--venue_year_review
      |--venue_year_id1_review.json
      |--venue_year_id2_review.json
      ...
   |--venue_year_pdf
      |--venue_year_id1_pdf.pdf
      |--venue_year_id2_pdf.pdf
      ...

Dataset Content

Paper Contents

• title: title of the paper
• authors: list of author names
• emails: list of author emails
• sections: list of sections of the paper
  • heading: heading of the section
  • text: text of the section
• references: list of references of the paper
  • title: title of the reference
  • author: list of author names of the reference
  • venue: venue of the reference
  • citeRegEx: citation expression
  • shortCiteRegEx: short citation expression
  • year: publication year of the reference
• referenceMentions: the location of the reference in the paper
  • referenceID: numerical reference id
  • context: context of the reference in the paper
  • startOffset: start index of the context
  • endOffset: end index of the context
• year: year of publication
• abstractText: abstract of the paper

Metadata Contents

• id: unique id of the paper
• conference: venue for the paper
• decision: final decision for the paper (accept/reject)
• url: link to the PDF of the paper
• review_url: link to the review of the paper
• title: title of the paper
• authors: list of the authors of the paper

Dataset Sources

We incorporate parts of the PeerRead and NLPeer datasets along with an update-to-date crawl from ICLR and NeurIPS on OpenReview and NeurIPS Proceedings.

Citation

If you find this dataset useful in your research, please cite the following paper:

@misc{gao2024reviewer2,
      title={Reviewer2: Optimizing Review Generation Through Prompt Generation}, 
      author={Zhaolin Gao and Kianté Brantley and Thorsten Joachims},
      year={2024},
      eprint={2402.10886},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}