KEEPHA-ADR Baselines: Pharmaceutical Information Extraction Systems for French, German, and Japanese.

---

Introduction

An adverse drug reaction (ADR) is a “harmful or unpleasant reaction, resulting from an intervention related to the use of a medicinal product” (Edwards and Aronson, 2000). Though clinical trials are conducted for every drug, the pool of trial participants can never represent the entire population. Thus, to actively monitor the effect of medical use, we built systems to extract online discussions about drug usage. The systems extract the following from a given piece of text.

• Name Entity Recognition (NER): identify l’infliximab as drug, toute rouge and nauséeuse as disorder.
• Attribute Classification (AC): classify the status of the drug l’infliximab as stopped.
• Relation Extraction (RE): realize caused relations between l’infliximab and toute rouge, between l’infliximab and nauséeuse.

Overall, the corpus contains annotations covering 12 entity types, 4 attribute types, and 13 relation types. The number of documents, the number of annotations for entities, relations, attributes per language:

Getting started

pip install -r ./requirements.txt

Data Access If you are interested in the data, please contact Lisa Raithel (raithel@tu-berlin.de) The dataset is to be extracted in the directory ./data

Example usage

To reproduce the results for the three Information Extraction tasks, you need a run file and a config file. You can find config files in the folder ./configs for respective tasks.

Name Entity Recognition (NER)

python src/finetune_ner.py configs/ner_configs/CONFIG_TRAIN.json
python src/inference_ner.py configs/eval_ner_configs/CONFIG_TEST.json

Attribute Classification (AC)

python src/finetune_att.py configs/att_configs/CONFIG_TRAIN.json
python src/inference_att.py configs/eval_att_configs/CONFIG_TEST.json

Relation Extraction (RE)

python src/finetune_re.py configs/re_configs/CONFIG_TRAIN.json
python src/inference_re.py configs/eval_re_configs/CONFIG_TEST.json

Evaluation

java -cp brateval.jar au.com.nicta.csp.brateval.CompareEntities EVAL_FOLDER REF_FOLDER false

Citation

Please cite our paper if you plan to use KEEPHA-ADR:

@inproceedings{raithel_dataset_2024,
  title = {A {{Dataset}} for {{Pharmacovigilance}} in {{German}}, {{French}}, and {{Japanese}}: {{Annotating Adverse Drug Reactions}} across {{Languages}}},
  booktitle = {Proceedings of the {{Language Resources}} and {{Evaluation Conference}}},
  author = {Raithel*, Lisa and Yeh*, Hui-Syuan and Yada, Shuntaro and Grouin, Cyril and Lavergne, Thomas and N{\'e}v{\'e}ol, Aur{\'e}lie and Paroubek, Patrick and Thomas, Philippe and M{\"o}ller, Sebastian and Nishiyama, Tomohiro and Aramaki, Eiji and Matsumoto, Yuji and Roller, Roland and Zweigenbaum, Pierre},
  year = {2024},
  month = may,
  publisher = {{ European Language Resources Association}},
  address = {{Torino, Italy}},
  copyright = {All rights reserved}
}