Datasets:

Languages:
English
Multilinguality:
monolingual
Size Categories:
100K<n<1M
Language Creators:
found
Annotations Creators:
found
Source Datasets:
original
ArXiv:
License:

`sections` not in dataset

#4
by acmc - opened

The original format of the dataset is as follows:

{ 
  'article_id': str,
  'abstract_text': List[str],
  'article_text': List[str],
  'section_names': List[str],
  'sections': List[List[str]]
}

However, the dataset on HF doesn't seem to contain the article text, separated by sections. Is there any workaround to obtain this, or shall I download the original dataset and process it on my own?

Datasets Maintainers org

Hi, @acmc , thanks for reaching out.

Yes, you are right. This dataset was originally ported from TensorFlow Datasets (https://www.tensorflow.org/datasets/catalog/scientific_papers) where the "sections" data (sentence-tokenized article text split by section) was also ignored.

I think the only way to obtain the section-split text is by accessing the original data files.

This is rather perplexing to hear... Was it ignored by TF Datasets because it would entail a repetition of the body of the article? That would be the only sensible explanation, but I don't think its an outright redundancy... After all, in its current form, it's impossible to have a clear notion of the boundaries of each part of the article.

Anyway, I'll rebuild the dataset on my own, thank you!

Sign up or log in to comment