configs:
- config_name: document_metadata
data_files:
- split: train
path: data/document_metadata/document_metadata_train.parquet
- split: dev
path: data/document_metadata/document_metadata_dev.parquet
- config_name: spans
data_files:
- split: train
path: data/spans/spans_train.parquet
- split: dev
path: data/spans/spans_dev.parquet
task_categories:
- token-classification
language:
- fr
license: cc-by-nc-4.0
tags:
- inception
- uima
- annotation
**Dataset Card for **
Dataset Description
Dataset Summary
is a subpart of the PARHAF corpus, an open French corpus of human-authored clinical reports of fictional patients.
It was created to support the development and evaluation of clinical NLP systems for .
This dataset contains training data only. The test set will remain under embargo to enable future evaluations under controlled conditions, limiting the risk of LLM contamination through prior data exposure. Please contact us for access to the test data.
Each patient record was:
- written by a senior medical resident
- reviewed by another senior medical resident, from the same specialty
- annotated by a specialist of the use case
- curated by another specialist of the use case
Data statistics
Indicateurs globaux
| Indicateur |
Valeur |
| Number of JSON files |
209 |
| Total annotations |
417 |
| Average document length |
2584 characters |
| 50% threshold (annotations) |
208 |
Distribution by type – complete dataset
| Type |
Annotations |
% of total |
| DocumentMetadata |
209 |
50.12% |
| Span |
208 |
49.88% |
Distribution by filled field – complete dataset
| Type.Field |
Occurrences |
% of total annot. |
| DocumentMetadata.Nomenclature |
209 |
50.12% |
| Span.Justification |
208 |
49.88% |
Train / Test Split Results
| Indicator |
TRAIN |
TOTAL |
| Number of files |
108 |
209 |
| Number of annotations |
211 |
417 |
| Dataset percentage |
50.60% |
100.00% |
| Avg. document length (chars) |
2573 |
|
Distribution by type – Train / Test
| Type |
TRAIN |
% train of type |
| DocumentMetadata |
108 |
51.67% |
| Span |
103 |
49.52% |
Distribution by filled field – Train / Test
| Type.Field |
TRAIN |
% train of field |
| DocumentMetadata.Nomenclature |
108 |
51.67% |
| Span.Justification |
103 |
49.52% |
Distribution by value – Train / Test
| Type.Field.Value |
TRAIN |
% train |
| Span.Justification.Justification |
103 |
49.52% |
| DocumentMetadata.Nomenclature.PartialResponse |
39 |
54.93% |
| DocumentMetadata.Nomenclature.ProgressiveDisease |
20 |
48.78% |
| DocumentMetadata.Nomenclature.StableDisease |
20 |
58.82% |
| DocumentMetadata.Nomenclature.NotApplicable |
15 |
48.39% |
| DocumentMetadata.Nomenclature.CompleteResponse |
11 |
44.00% |
| DocumentMetadata.Nomenclature.Undetermined |
3 |
42.86% |
Data Origin
The clinical reports are extracted from the PARHAF corpus. Please refer to PARHAF documentation for more information about this corpus.
Languages
Dataset Structure
We distribute both a Hugging Face dataset and a standalone version of the corpus. The standalone dataset consists of a JSON file per patient report, in UIMA CAS JSON format. This format constitutes the canonical version of the corpus. The Hugging Face dataset (Parquet/Arrow) is a derived representation generated automatically from the JSON files.
Both formats therefore contain identical information and differ only in storage layout.
One dataset instance corresponds to one report.
Data Fields
| Champ |
Type |
Description |
Valeurs possibles |
| document_metadata |
|
|
|
report |
string |
Identifiant unique du rapport |
|
full_text |
string |
Texte integral du rapport |
|
| spans |
|
|
|
report |
string |
Identifiant du rapport |
|
span_id |
integer |
Identifiant de l'annotation |
|
span_type |
string |
Type de l'entite annotee |
Span |
begin |
integer |
Offset de debut |
|
end |
integer |
Offset de fin |
|
span_text |
string |
Texte de l'entite |
|
attribute_Justification |
string |
Justification |
Justification |
Data Splits
Only the training set is released here. The remaining portion of the corpus will be temporarily embargoed to enable future evaluations under controlled conditions, thereby limiting the risk of large language model contamination through prior exposure to the data. You can evaluate your system on the test set through the CodaBench platform.
Additional Information
Licensing Information
This dataset is released under licenses:
Citation Information
[More Information Needed]
