SoFairOA/softcite_dataset

Softcite Dataset

Softcite dataset is a dataset of software mentions in scientific papers. This is the second version of this dataset that is originally available at https://github.com/softcite/softcite_dataset_v2.

In contrast to the dataset at GitHUB repository, this dataset contains the full text of the papers also for the train set. Originally, the full text was only available for the test set.

Empty documents were filtered out.

Splits naming

• train: The training split corresponds to the softcite_corpus-working in the original dataset.
• test: The test split corresponds to the softcite_corpus-holdout in the original dataset.

Configurations

This dataset is available in two configurations:

• paragraphs: This configuration contains the dataset in a format where each paragraph is a separate example. This configuration is the default one.
• documents: This configuration contains the dataset in a format where each document is a separate example.

Fields description

Both configurations contain the following fields:

• id: The unique identifier of the example. In case of paragraphs configuration, the id format is {paper_id}_{paragraph_index}.
• title: The title of the paper.
• doi: The DOI of the paper if available.
• text: The text of the paragraph or the full text of the paper.
• tokens: Tokenized text for which the labels are provided.
• start_offsets: The start character offsets of the tokens in the text.
• labels: The labels of the tokens. The labels are in the BIO format:
  • 0: O
  • 1: B-SOFTWARE
  • 2: I-SOFTWARE
  • 3: B-VERSION
  • 4: I-VERSION
  • 5: B-PUBLISHER
  • 6: I-PUBLISHER
  • 7: B-URL
  • 8: I-URL
  • 9: B-LANGUAGE
  • 10: I-LANGUAGE
• software: This field is only available in the documents configuration. It contains the following fields:
  • name: The name of the software.
  • version: The versions of the software.
  • publisher: The publishers of the software.
  • url: The URLs of the software.
  • language: The programming languages of the software.

Disambiguation is done on lexical level. Meaning that two software names are considered different if they have different normalized forms.

License

This dataset is ment only for internal usage within the SoFAIR project.

The original upload is distributed under CC-BY license. The Python scrips are distributed under Apache 2.0 license.