Integrating higher-level semantics into robust biomedical name representations

This directory contains source code for the following paper:

Integrating Higher-Level Semantics into Robust Biomedical Name Representations.
Pieter Fivez, Simon Šuster and Walter Daelemans. LOUHI (EACL), 2021.

If you use this code, please cite:

@inproceedings{fivez-etal-2021-integrating,
title = "Integrating Higher-Level Semantics into Robust Biomedical Name Representations",
author = "Fivez, Pieter  and
  Suster, Simon  and
  Daelemans, Walter",
booktitle = "Proceedings of the 12th International Workshop on Health Text Mining and Information Analysis",
year = "2021",
publisher = "Association for Computational Linguistics",
pages = "49--58"}

License

GPL-3.0

Requirements

All requirements are listed in requirements.txt.

You can run pip install -r requirements.txt, preferably in a virtual environment.

The fastText model used in the paper can be downloaded from the following link: https://drive.google.com/file/d/1B07lc3eeW_zughHguugLBR4iJYQj_Wxz/view?usp=sharing
Our example script requires a path to this downloaded model.

Data

Since we are not allowed to share SNOMED-CT data, we demonstrate our code using the openly available MedMentions corpus.
We have used this corpus as fine-grained synonym sets in this previous publication:

@inproceedings{fivez-etal-2021-conceptual,
title = "Conceptual Grounding Constraints for Truly Robust Biomedical Name Representations",
author = "Fivez, Pieter  and
  Suster, Simon  and
  Daelemans, Walter",
booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume",
year = "2021",
publisher = "Association for Computational Linguistics",
pages = "2440--2450"}

The source files for this corpus can be found at https://github.com/chanzuckerberg/MedMentions.

The script data/extract_medmentions.py has used these source files to create data/medmentions.json.

Code

We provide a script to run our training objectives from the paper.

main_dan.py trains and evaluates the DAN encoder on data/medmentions.json.
Please run python main_dan.py --help to see the options, or check the script.
The default parameters are the best parameters reported in our paper.