README.md

Dataset

• split_survey_df: The split version of SciReviewGen, which aims to generate literature review chapters
• original_survey_df: The original version of SciReviewGen, which aims to generate the entire text of literature reviews
• summarization_csv: CSV files suitable for summarization task.

Data format

split_survey_df & original_survey_df

• Row:
  • literature review chapter or the entire text of literature review
• Column:
  • paper_id: paper_id used in S2ORC
  • title: title of the literature review
  • abstract: abstract of the literature review
  • section: chapter title
  • text: body text of literature review chapter or literature review paper
  • n_bibs: number of the cited papers that can be used as inputs
  • n_nonbibs: number of the cited papers that cannot be used as inputs
  • bib_titles: titles of the cited papers
  • bib_abstracts: abstracts of the cited papers
  • bib_citing_sentences: citing sentences that cite the cited papers
  • split: train/val/test split

summarization_csv

• Row:
  • literature review chapter
• Column:
  • reference: literature review title <s> chapter title <s> abstract of cited paper 1 <s> BIB001 </s> literature review title <s> chapter title <s> abstract of cited paper 2 <s> BIB002 </s> ...
  • target: literature review chapter

Data construction

Use make_summarization_csv.py to convert df to csv

Reference

• https://github.com/tetsu9923/SciReviewGen