We present NERmembert-large-4entities, which is a CamemBERT large fine-tuned for the Name Entity Recognition task for the French language on four French NER datasets for 4 entities (LOC, PER, ORG, MISC).
All these datasets were concatenated and cleaned into a single dataset that we called frenchNER_4entities.
There are a total of 384,773 rows, of which 328,757 are for training, 24,131 for validation and 31,885 for testing.
Our methodology is described in a blog post available in English or French.

Dataset

The dataset used is frenchNER_4entities, which represents ~385k sentences labeled in 4 categories:

Label	Examples
PER	"La Bruyère", "Gaspard de Coligny", "Wittgenstein"
ORG	"UTBM", "American Airlines", "id Software"
LOC	"République du Cap-Vert", "Créteil", "Bordeaux"
MISC	"Wolfenstein 3D", "Révolution française", "Coupe du monde de football"

The distribution of the entities is as follows:

Splits	O	PER	LOC	ORG	MISC
train	7,539,692	307,144	286,746	127,089	799,494
validation	544,580	24,034	21,585	5,927	18,221
test	720,623	32,870	29,683	7,911	21,760

Evaluation results

The evaluation was carried out using the evaluate python package.

frenchNER_4entities

For space reasons, we show only the F1 of the different models. You can see the full results below the table.

Model	PER	LOC	ORG	MISC
Jean-Baptiste/camembert-ner	0.971	0.947	0.902	0.663
cmarkea/distilcamembert-base-ner	0.974	0.948	0.892	0.658
NERmembert-base-3entities	0.978	0.957	0.904	0
NERmembert-large-3entities	0.980	0.963	0.919	0
NERmembert-base-4entities	0.978	0.958	0.903	0.814
NERmembert-large-4entities (this model)	0.982	0.964	0.919	0.834

Full results

Model	Metrics	PER	LOC	ORG	MISC	O	Overall
Jean-Baptiste/camembert-ner	Precision	0.952	0.924	0.870	0.845	0.986	0.976
Recall	0.990	0.972	0.938	0.546	0.992	0.976
F1	0.971	0.947	0.902	0.663	0.989	0.976
cmarkea/distilcamembert-base-ner	Precision	0.962	0.933	0.857	0.830	0.985	0.976
Recall	0.987	0.963	0.930	0.545	0.993	0.976
F1	0.974	0.948	0.892	0.658	0.989	0.976
NERmembert-base-3entities	Precision	0.973	0.955	0.886	0	X	X
Recall	0.983	0.960	0.923	0	X	X
F1	0.978	0.957	0.904	0	X	X
NERmembert-large-3entities	Precision	0.978	0.960	0.899	0	X	X
Recall	0.985	0.966	0.940	0	X	X
F1	0.980	0.963	0.919	0	X	X
NERmembert-base-4entities	Precision	0.973	0.951	0.888	0.850	0.993	0.984
Recall	0.983	0.964	0.918	0.781	0.993	0.984
F1	0.978	0.958	0.903	0.814	0.993	0.984
NERmembert-large-4entities (this model)	Precision	0.977	0.961	0.896	0.872	0.993	0.986
Recall	0.987	0.966	0.943	0.798	0.995	0.986
F1	0.982	0.964	0.919	0.834	0.994	0.986

In detail:

multiconer

For space reasons, we show only the F1 of the different models. You can see the full results below the table.

Model	PER	LOC	ORG	MISC
Jean-Baptiste/camembert-ner	0.940	0.761	0.723	0.560
cmarkea/distilcamembert-base-ner	0.921	0.748	0.694	0.530
NERmembert-base-3entities	0.960	0.887	0.877	0
NERmembert-large-3entities	0.965	0.902	0.896	0
NERmembert-base-4entities	0.960	0.890	0.867	0.852
NERmembert-large-4entities (this model)	0.969	0.919	0.904	0.864

Full results

Model	Metrics	PER	LOC	ORG	MISC	O	Overall
Jean-Baptiste/camembert-ner	Precision	0.908	0.717	0.753	0.620	0.936	0.889
Recall	0.975	0.811	0.696	0.511	0.938	0.889
F1	0.940	0.761	0.723	0.560	0.937	0.889
cmarkea/distilcamembert-base-ner	Precision	0.885	0.738	0.737	0.589	0.928	0.881
Recall	0.960	0.759	0.655	0.482	0.939	0.881
F1	0.921	0.748	0.694	0.530	0.934	0.881
NERmembert-base-3entities	Precision	0.957	0.894	0.876	0	X	X
Recall	0.962	0.880	0.878	0	X	X
F1	0.960	0.887	0.877	0	X	X
NERmembert-large-3entities	Precision	0.960	0.903	0.916	0	X	X
Recall	0.970	0.900	0.877	0	X	X
F1	0.965	0.902	0.896	0	X	X
NERmembert-base-4entities	Precision	0.954	0.893	0.851	0.849	0.979	0.954
Recall	0.967	0.887	0.883	0.855	0.974	0.954
F1	0.960	0.890	0.867	0.852	0.977	0.954
NERmembert-large-4entities (this model)	Precision	0.964	0.922	0.904	0.856	0.981	0.961
Recall	0.975	0.917	0.904	0.872	0.976	0.961
F1	0.969	0.919	0.904	0.864	0.978	0.961

multinerd

For space reasons, we show only the F1 of the different models. You can see the full results below the table.

Model	PER	LOC	ORG	MISC
Jean-Baptiste/camembert-ner	0.962	0.934	0.888	0.419
cmarkea/distilcamembert-base-ner	0.972	0.938	0.884	0.430
NERmembert-base-3entities	0.985	0.973	0.938	0
NERmembert-large-3entities	0.987	0.979	0.953	0
NERmembert-base-4entities	0.985	0.973	0.938	0.770
NERmembert-large-4entities (this model)	0.987	0.976	0.948	0.790

Full results

Model	Metrics	PER	LOC	ORG	MISC	O	Overall
Jean-Baptiste/camembert-ner	Precision	0.931	0.893	0.827	0.725	0.979	0.966
Recall	0.994	0.980	0.959	0.295	0.990	0.966
F1	0.962	0.934	0.888	0.419	0.984	0.966
cmarkea/distilcamembert-base-ner	Precision	0.954	0.908	0.817	0.705	0.977	0.967
Recall	0.991	0.969	0.963	0.310	0.990	0.967
F1	0.972	0.938	0.884	0.430	0.984	0.967
NERmembert-base-3entities	Precision	0.974	0.965	0.910	0	X	X
Recall	0.995	0.981	0.968	0	X	X
F1	0.985	0.973	0.938	0	X	X
NERmembert-large-3entities	Precision	0.979	0.970	0.927	0	X	X
Recall	0.996	0.987	0.980	0	X	X
F1	0.987	0.979	0.953	0	X	X
NERmembert-base-4entities	Precision	0.976	0.961	0.911	0.829	0.991	0.983
Recall	0.994	0.985	0.967	0.719	0.993	0.983
F1	0.985	0.973	0.938	0.770	0.992	0.983
NERmembert-large-4entities (this model)	Precision	0.979	0.967	0.922	0.852	0.991	0.985
Recall	0.996	0.986	0.974	0.736	0.994	0.985
F1	0.987	0.976	0.948	0.790	0.993	0.985

wikiner

For space reasons, we show only the F1 of the different models. You can see the full results below the table.

Model	PER	LOC	ORG	MISC
Jean-Baptiste/camembert-ner	0.986	0.966	0.938	0.938
cmarkea/distilcamembert-base-ner	0.983	0.964	0.925	0.926
NERmembert-base-3entities	0.970	0.945	0.878	0
NERmembert-large-3entities	0.972	0.953	0.893	0
NERmembert-base-4entities	0.970	0.945	0.876	0.872
NERmembert-large-4entities (this model)	0.975	0.953	0.896	0.893

Full results

Model	Metrics	PER	LOC	ORG	MISC	O	Overall
Jean-Baptiste/camembert-ner	Precision	0.986	0.962	0.925	0.943	0.998	0.992
Recall	0.987	0.969	0.951	0.933	0.997	0.992
F1	0.986	0.966	0.938	0.938	0.998	0.992
cmarkea/distilcamembert-base-ner	Precision	0.982	0.964	0.910	0.942	0.997	0.991
Recall	0.985	0.963	0.940	0.910	0.998	0.991
F1	0.983	0.964	0.925	0.926	0.997	0.991
NERmembert-base-3entities	Precision	0.971	0.947	0.866	0	X	X
Recall	0.969	0.943	0.891	0	X	X
F1	0.970	0.945	0.878	0	X	X
NERmembert-large-3entities	Precision	0.973	0.953	0.873	0	X	X
Recall	0.971	0.948	0.913	0	X	X
F1	0.972	0.953	0.893	0	X	X
NERmembert-base-4entities	Precision	0.970	0.944	0.872	0.878	0.996	0.986
Recall	0.969	0.947	0.880	0.866	0.996	0.986
F1	0.970	0.945	0.876	0.872	0.996	0.986
NERmembert-large-4entities (this model)	Precision	0.975	0.957	0.872	0.901	0.997	0.989
Recall	0.975	0.949	0.922	0.884	0.997	0.989
F1	0.975	0.953	0.896	0.893	0.997	0.989

Usage

Code

python

from transformers import pipeline

ner = pipeline('token-classification', model='CATIE-AQ/NERmembert-large-4entities', tokenizer='CATIE-AQ/NERmembert-large-4entities', aggregation_strategy="simple")

results = ner(
"Le dévoilement du logo officiel des JO s'est déroulé le 21 octobre 2019 au Grand Rex. Ce nouvel emblème et cette nouvelle typographie ont été conçus par le designer Sylvain Boyer avec les agences Royalties & Ecobranding. Rond, il rassemble trois symboles : une médaille d'or, la flamme olympique et Marianne, symbolisée par un visage de femme mais privée de son bonnet phrygien caractéristique. La typographie dessinée fait référence à l'Art déco, mouvement artistique des années 1920, décennie pendant laquelle ont eu lieu pour la dernière fois les Jeux olympiques à Paris en 1924. Pour la première fois, ce logo sera unique pour les Jeux olympiques et les Jeux paralympiques."
)

print(result)

python

[{'entity_group': 'MISC', 'score': 0.9922348, 'word': 'JO', 'start': 35, 'end': 38},
{'entity_group': 'LOC', 'score': 0.9995632, 'word': 'Grand Rex', 'start': 74, 'end': 84},
{'entity_group': 'PER', 'score': 0.99612623, 'word': 'Sylvain Boyer', 'start': 164, 'end': 178},
{'entity_group': 'ORG', 'score': 0.87376696, 'word': 'Royalties & Ecobranding', 'start': 195, 'end': 219},
{'entity_group': 'PER', 'score': 0.8121169, 'word': 'Marianne', 'start': 298, 'end': 307},
{'entity_group': 'MISC', 'score': 0.7016645, 'word': 'Art déco', 'start': 438, 'end': 446},
{'entity_group': 'MISC', 'score': 0.99880475, 'word': 'Jeux olympiques', 'start': 549, 'end': 565},
{'entity_group': 'LOC', 'score': 0.98473144, 'word': 'Paris', 'start': 567, 'end': 573},
{'entity_group': 'MISC', 'score': 0.99421215, 'word': 'Jeux olympiques', 'start': 634, 'end': 650},
{'entity_group': 'MISC', 'score': 0.9936283, 'word': 'Jeux paralympiques', 'start': 657, 'end': 676}]

Try it through Space

A Space has been created to test the model. It is available here.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.0347	1.0	41095	0.0537	0.9832	0.9832	0.9832	0.9832
0.0237	2.0	82190	0.0448	0.9858	0.9858	0.9858	0.9858
0.0119	3.0	123285	0.0532	0.9860	0.9860	0.9860	0.9860

Framework versions

Transformers 4.36.2
Pytorch 2.1.2
Datasets 2.16.1
Tokenizers 0.15.0

Environmental Impact

Carbon emissions were estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). The hardware, runtime, cloud provider, and compute region were utilized to estimate the carbon impact.

Hardware Type: A100 PCIe 40/80GB
Hours used: 4h17min
Cloud Provider: Private Infrastructure
Carbon Efficiency (kg/kWh): 0.078 (estimated from electricitymaps for the day of January 10, 2024.)
Carbon Emitted (Power consumption x Time x Carbon produced based on location of power grid): 0.08 kg eq. CO2

Citations

NERmembert-large-4entities

@misc {NERmembert2024,
    author       = { {BOURDOIS, Loïck} },  
    organization = { {Centre Aquitain des Technologies de l'Information et Electroniques} },  
	title        = { NERmembert-large-4entities (Revision 1cd8be5) },
	year         = 2024,
	url          = { https://huggingface.co/CATIE-AQ/NERmembert-large-4entities },
	doi          = { 10.57967/hf/1752 },
	publisher    = { Hugging Face }
}

multiconer

@inproceedings{multiconer2-report,  
    title={{SemEval-2023 Task 2: Fine-grained Multilingual Named Entity Recognition (MultiCoNER 2)}},  
    author={Fetahu, Besnik and Kar, Sudipta and Chen, Zhiyu and Rokhlenko, Oleg and Malmasi, Shervin},  
    booktitle={Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023)},  
    year={2023},  
    publisher={Association for Computational Linguistics}}

@article{multiconer2-data,  
    title={{MultiCoNER v2: a Large Multilingual dataset for Fine-grained and Noisy Named Entity Recognition}},  
    author={Fetahu, Besnik and Chen, Zhiyu and Kar, Sudipta and Rokhlenko, Oleg and Malmasi, Shervin},  
    year={2023}}

multinerd

 @inproceedings{tedeschi-navigli-2022-multinerd,  
    title = "{M}ulti{NERD}: A Multilingual, Multi-Genre and Fine-Grained Dataset for Named Entity Recognition (and Disambiguation)",  
    author = "Tedeschi, Simone and  Navigli, Roberto",  
    booktitle = "Findings of the Association for Computational Linguistics: NAACL 2022",  
    month = jul,  
    year = "2022",  
    address = "Seattle, United States",  
    publisher = "Association for Computational Linguistics",  
    url = "https://aclanthology.org/2022.findings-naacl.60",  
    doi = "10.18653/v1/2022.findings-naacl.60",  
    pages = "801--812"}

pii-masking-200k

@misc {ai4privacy_2023,  
    author = { {ai4Privacy} },  
    title = { pii-masking-200k (Revision 1d4c0a1) },  
    year = 2023,  
    url = { https://huggingface.co/datasets/ai4privacy/pii-masking-200k },  
    doi = { 10.57967/hf/1532 },  
    publisher = { Hugging Face }}

wikiann

@inproceedings{rahimi-etal-2019-massively,  
    title = "Massively Multilingual Transfer for {NER}",  
    author = "Rahimi, Afshin and Li, Yuan and Cohn, Trevor",  
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",  
    month = jul,  
    year = "2019",  
    address = "Florence, Italy",  
    publisher = "Association for Computational Linguistics",  
    url = "https://www.aclweb.org/anthology/P19-1015",  
    pages = "151--164"}

wikiner

@article{NOTHMAN2013151,  
    title = {Learning multilingual named entity recognition from Wikipedia},  
    journal = {Artificial Intelligence},  
    volume = {194},  
    pages = {151-175},  
    year = {2013},  
    note = {Artificial Intelligence, Wikipedia and Semi-Structured Resources},  
    issn = {0004-3702},  
    doi = {https://doi.org/10.1016/j.artint.2012.03.006},  
    url = {https://www.sciencedirect.com/science/article/pii/S0004370212000276},  
    author = {Joel Nothman and Nicky Ringland and Will Radford and Tara Murphy and James R. Curran}}

frenchNER_4entities

@misc {frenchNER2024,  
    author       = { {BOURDOIS, Loïck} },  
    organization  = { {Centre Aquitain des Technologies de l'Information et Electroniques} },  
    title        = { frenchNER_4entities },  
    year         = 2024,  
    url          = { https://huggingface.co/CATIE-AQ/frenchNER_4entities },  
    doi          = { 10.57967/hf/1751 },  
    publisher    = { Hugging Face }  
}

CamemBERT

@inproceedings{martin2020camembert,  
  title={CamemBERT: a Tasty French Language Model},  
  author={Martin, Louis and Muller, Benjamin and Su{\'a}rez, Pedro Javier Ortiz and Dupont, Yoann and Romary, Laurent and de la Clergerie, {\'E}ric Villemonte and Seddah, Djam{\'e} and Sagot, Beno{\^\i}t},  
  booktitle={Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},  
  year={2020}}

License

MIT