Dataset Card for MedicalNER_Fr

The dataset MedicalNER_Fr has been specifically curated to facilitate training Named Entity Recognition (NER) models for the French language within the medical and healthcare domain. It is derived from the Multilingual Complex Named Entity Recognition (MultiCoNER) Dataset and is intended solely for educational purposes.

The MultiCoNER V2 dataset has undergone filtration to exclusively encompass French-language entries associated with the medical domain. Non-medical tags have been aggregated into broader categories. Before commencing the training of your NER medical model, it is advisable to address the imbalanced nature of the dataset to ensure optimal training outcomes.

Dataset Details

Dataset Description

• Curated by: typica.ai
• License: cc-by-4.0

Uses

The dataset is designed to train Named Entity Recognition models for the French language in the medical and healthcare domain.

Dataset Structure

1. sample_id: A UUID generated for each example.
2. tokens: A list of tokens (words) in the sentence.
3. ner_tags: A list of named entity recognition (NER) tags corresponding to each token. These tags indicate the entity type of each token.
4. text: Text formed by combining the tokens.
5. ner_tags_span: A list of spans for the NER tags. Each span is a list containing:
   • The NER tag (entity type).
   • The start position of the entity in the text.
   • The end position of the entity in the text.

Dataset Tags Count:

• AnatomicalStructure: 4685
• Disease: 4658
• Medication/Vaccine: 4226
• MedicalProcedure: 3170
• Symptom: 1763
• LOC: 525
• PER: 521
• PROD: 305
• CW: 167
• ORG: 83
• GRP: 14

Example

{'sample_id': '60a82e36-4d34-4e16-aadc-2078699476f7',
 'tokens': ['jonas',
  'salk',
  'médecin',
  'm.d.',
  '1938',
  'et',
  'inventeur',
  'du',
  'vaccin',
  'contre',
  'la',
  'poliomyélite',
  '.'],
 'ner_tags': ['B-PER',
  'I-PER',
  'O',
  'O',
  'O',
  'O',
  'O',
  'O',
  'O',
  'O',
  'O',
  'B-Disease',
  'O'],
 'text': 'jonas salk médecin m.d. 1938 et inventeur du vaccin contre la poliomyélite .',
 'ner_tags_span': "[['PER', 0, 10], ['Disease', 62, 74]]"}

Dataset Creation

Curation Rationale

This dataset was created for educational purposes only.

Source Data

The Dataset source is Multilingual Complex Named Entity Recognition (MultiCoNER V2).

Data Collection and Processing

The MultiCoNER V2 dataset has been filtered to include only French language rows and only those related to the medical domain. Non-medical tags have been aggregated into coarse-grained tags.

Bias, Risks, and Limitations

This dataset was created for educational purposes only.

Recommendations

To ensure optimal training for your NER medical model, it is recommended to balance the unbalanced dataset before proceeding.

Citation

If you use this dataset, please cite:

@misc{MedicalNER_Fr2024,
  author = {Hicham Assoudi},
  title = {MedicalNER_Fr: Named Entity Recognition Dataset for the French language in the medical and healthcare domain},
  note = {Created by Hicham Assoudi, Ph.D. at Typica.ai (url{https://typica.ai/}), published on Hugging Face},
  year = {2024},
  url = {https://huggingface.co/datasets/TypicaAI/MedicalNER_Fr}
}

Dataset Contact

Feel free to reach out to us at assoudi@typica.ai if you have any questions or comments.