Skip to content

Latest commit

 

History

History
13 lines (9 loc) · 867 Bytes

README.md

File metadata and controls

13 lines (9 loc) · 867 Bytes
Raw

Datasets

For training, use columns text and label. label is a numerical label with the following mapping:

num2label = {0: "created", 1: "used", 2: "mention", 3: "none"}
label2num = {"created": 0, "used": 1, "mention": 2, "none": 3}

To regenerate the CSV files from the Excel sheets, run the notebook load_data.ipynb.

  • software_citation_intent_merged.csv: dataset build by Josh, Martin and Kai from Softcite, SoMeSci and unlabelled examples. Also contains a column context which has the surrounding sentences. In the case of the unlabelled examples, this is currently equivalent to the sentences column but that will be updated soon.
  • software_citation_intent_czi.csv: dataset curated by CZI, provided by Ana and mapped to our intent labels by Donghui. Does not contain context column. Removed rows with unclear/NaN citation intent.