metadata
language: br
task_categories:
- token-classification
configs:
- config_name: avec_prefixe
data_files:
- split: train
path: data/avec_prefixe/train-*
- split: validation
path: data/avec_prefixe/validation-*
- split: test
path: data/avec_prefixe/test-*
- config_name: sans_prefixe
data_files:
- split: train
path: data/sans_prefixe/train-*
- split: validation
path: data/sans_prefixe/validation-*
- split: test
path: data/sans_prefixe/test-*
Description
Version nettoyée de WikiAnn.
En effet, la version originale contenait des leaks et des duplications.
De 1000 effectifs par split, la nouvelle répartition devient alors la suivante :
DatasetDict({
train: Dataset({
features: ['tokens', 'ner_tags'],
num_rows: 915
})
validation: Dataset({
features: ['tokens', 'ner_tags'],
num_rows: 946
})
test: Dataset({
features: ['tokens', 'ner_tags'],
num_rows: 952
})
})
Configurations
Le jeu de données est disponible en deux configurations avec prefixe (i.e. schéma IOB) et sans préfixe.
Avec prefixe
Le dictionnaire label2id est disponible ici.
{
"B-LOC": 0,
"B-ORG": 1,
"B-PER": 2,
"I-LOC": 3,
"I-ORG": 4,
"I-PER": 5,
"O": 6
}
La répartition des entités est alors la suivante
O | I-ORG | B-LOC | B-ORG | B-PER | I-LOC | I-PER | |
---|---|---|---|---|---|---|---|
train | 4502 | 525 | 360 | 710 | 284 | 592 | 280 |
validation | 4179 | 427 | 396 | 767 | 283 | 581 | 292 |
test | 4491 | 505 | 336 | 639 | 348 | 808 | 345 |
Sans prefixe
Le dictionnaire label2id est alors
{
"O": 0,
"PER": 1,
"ORG": 2,
"LOC": 3
}
Et la répartition des entités devient
O | LOC | PER | ORG | |
---|---|---|---|---|
train | 4502 | 805 | 1070 | 876 |
validation | 4179 | 719 | 1163 | 864 |
test | 4491 | 850 | 975 | 1156 |