arXiv_abstract / README.md
LovelyBuggies's picture
Update README.md
eac142a verified
metadata
dataset_info:
  features:
    - name: article_id
      dtype: string
    - name: abstract_text
      dtype: string
    - name: token_count
      dtype: int64
  splits:
    - name: train
      num_bytes: 150590869
      num_examples: 140313
    - name: test
      num_bytes: 5848235
      num_examples: 5481
    - name: val
      num_bytes: 5748332
      num_examples: 5383
  download_size: 90308446
  dataset_size: 162187436
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: val
        path: data/val-*

arXiv Abstract

This dataset is based on the arXiv scientific papers and is used for the text expansion task. (Download raw data here).

I processed the raw data for the article expansion task with extract_arXiv_abstract.py. The processed dataset only contains the article ID and abstract fields, and the abstract length should be 100-300 tokens. The JSON objects are in the following format:

{ 
  'article_id': str,
  'abstract_text': List[str],
  'token_count': int
}