ContributionSum

The ContributionSum Dataset

Dataset Format

contribution.json

Each paper is represented as a key-value pair where the key is its arxiv id and the value contains the following fileds

• structure: list of section titles (and its subsection titles if presented)
• title: title of the paper
• abstract: abstract of the paper
• prompt: detected contribution prompt
• extracted_contribution: list of author-written contributions
• contribution_types: types of extracted contributions based on our annotation scheme: 0 - Approach, 1 - Result, 2 - Analysis, 3 - Topic or Resource
• SECTION_TITLE: full text from a specific section

split.json

Includes train, test and validation set

Shared drive

https://drive.google.com/drive/folders/1p51KpJWBB-hV4PlF8Ms8qFXn1lhhKRKV?usp=drive_link