S1000 corpus

The S1000 corpus is a comprehensive manual reannotation and extension of the S800 corpus that allows highly accurate recognition of species names, both for machine learning and dictionary-based methods. In this page we have gathered links to the publication, the corpus, datasets and codebases related to this project.

Publication and Corpus

• Publication in Oxford Bioinformatics: S1000: A better taxonomic name corpus for biomedical information extraction

• The S1000 Corpus in BRAT standoff format

Datasets

• The Zenodo project related to S1000 that contains:
  • The S1000 corpus split in training, development and test sets in BRAT and CoNLL formats
  • The guidelines used during annotation of the corpus (also available as an annodoc here)
  • the dictionary used by Jensenlab tagger
  • results from large scale tagging with the Jensenlab tagger
  • the model used for the large scale run of the transformer-based method and
  • results from large scale tagging with the transformer-based method
• The input documents for the large scale runs with:
  • Jensenlab tagger are hosted here and here
  • the transformer-based method are hosted here (Note: the pre-processed input documents are the same, the difference is in the document format for the two methods)

Code

• Useful scripts to reproduce the results presented in the publication can be found in this Github repo

• The codebase for the transformer-based NER tagger can be found here and the codebase used for the large scale run here