/CLEFeHealth2020-multilabel-bert

This repository includes the source code used in nlp4life@clef2020eHealth and fine-tuning BERT for multilabel sequence classification using HuggingFace Transformer.

Primary LanguagePythonMIT LicenseMIT

clef2020eHealth-multilabel-bert

This repository includes the source code used in nlp4life@clef2020eHealth and fine-tuning BERT for multilabel sequence classification using HuggingFace Transformer.

Requirements

Data format

All input data must be in csv or tsv. Content of the input files must include filename, text and one column per ICD code to indicate weather the ICD code is associate to the corresponding filename or not. For example, assume in general we have three different ICD codes in our ground truth set:

filename text g45.9 i08.1 i72.2
file1 text1 1 0 1

This means that file1 is labeled with ICD10 codes g45.9 and i72.2.

Python dependencies

Dependcies exist in the requirements.txt file. Simply install them by pip install -r requirements.txt

Data

  • Pretrained fastText embedding from SBWC can be downloaded from here.
  • The dataset from CLEFeHealth2020 challenge can be found at here.