This is a SpanMarker model trained on the FewNERD dataset that can be used for Named Entity Recognition. This SpanMarker model uses bert-base-multilingual-cased as the underlying encoder.

Model Details

Model Description

Model Type: SpanMarker
Encoder: bert-base-multilingual-cased
Maximum Sequence Length: 256 tokens
Maximum Entity Length: 8 words
Training Dataset: FewNERD
Languages: en, multilingual
License: cc-by-sa-4.0

Model Sources

Repository: SpanMarker on GitHub
Thesis: SpanMarker For Named Entity Recognition

Model Labels

Label	Examples
art-broadcastprogram	"Corazones", "Street Cents", "The Gale Storm Show : Oh , Susanna"
art-film	"L'Atlantide", "Bosch", "Shawshank Redemption"
art-music	"Atkinson , Danko and Ford ( with Brockie and Hilton )", "Hollywood Studio Symphony", "Champion Lover"
art-other	"Aphrodite of Milos", "The Today Show", "Venus de Milo"
art-painting	"Production/Reproduction", "Touit", "Cofiwch Dryweryn"
art-writtenart	"The Seven Year Itch", "Time", "Imelda de ' Lambertazzi"
building-airport	"Luton Airport", "Newark Liberty International Airport", "Sheremetyevo International Airport"
building-hospital	"Hokkaido University Hospital", "Yeungnam University Hospital", "Memorial Sloan-Kettering Cancer Center"
building-hotel	"Flamingo Hotel", "The Standard Hotel", "Radisson Blu Sea Plaza Hotel"
building-library	"British Library", "Bayerische Staatsbibliothek", "Berlin State Library"
building-other	"Communiplex", "Henry Ford Museum", "Alpha Recording Studios"
building-restaurant	"Fatburger", "Carnegie Deli", "Trumbull"
building-sportsfacility	"Sports Center", "Glenn Warner Soccer Facility", "Boston Garden"
building-theater	"Sanders Theatre", "Pittsburgh Civic Light Opera", "National Paris Opera"
event-attack/battle/war/militaryconflict	"Vietnam War", "Jurist", "Easter Offensive"
event-disaster	"1693 Sicily earthquake", "the 1912 North Mount Lyell Disaster", "1990s North Korean famine"
event-election	"March 1898 elections", "1982 Mitcham and Morden by-election", "Elections to the European Parliament"
event-other	"Eastwood Scoring Stage", "Masaryk Democratic Movement", "Union for a Popular Movement"
event-protest	"Russian Revolution", "Iranian Constitutional Revolution", "French Revolution"
event-sportsevent	"Stanley Cup", "World Cup", "National Champions"
location-GPE	"Mediterranean Basin", "Croatian", "the Republic of Croatia"
location-bodiesofwater	"Norfolk coast", "Atatürk Dam Lake", "Arthur Kill"
location-island	"Staten Island", "Laccadives", "new Samsat district"
location-mountain	"Miteirya Ridge", "Ruweisat Ridge", "Salamander Glacier"
location-other	"Victoria line", "Cartuther", "Northern City Line"
location-park	"Painted Desert Community Complex Historic District", "Shenandoah National Park", "Gramercy Park"
location-road/railway/highway/transit	"Friern Barnet Road", "Newark-Elizabeth Rail Link", "NJT"
organization-company	"Church 's Chicken", "Dixy Chicken", "Texas Chicken"
organization-education	"MIT", "Barnard College", "Belfast Royal Academy and the Ulster College of Physical Education"
organization-government/governmentagency	"Supreme Court", "Diet", "Congregazione dei Nobili"
organization-media/newspaper	"TimeOut Melbourne", "Clash", "Al Jazeera"
organization-other	"IAEA", "Defence Sector C", "4th Army"
organization-politicalparty	"Al Wafa ' Islamic", "Kenseitō", "Shimpotō"
organization-religion	"Christian", "UPCUSA", "Jewish"
organization-showorganization	"Lizzy", "Mr. Mister", "Bochumer Symphoniker"
organization-sportsleague	"China League One", "NHL", "First Division"
organization-sportsteam	"Luc Alphand Aventures", "Tottenham", "Arsenal"
other-astronomything	"`` Caput Larvae ''", "Algol", "Zodiac"
other-award	"GCON", "Order of the Republic of Guinea and Nigeria", "Grand Commander of the Order of the Niger"
other-biologything	"BAR", "Amphiphysin", "N-terminal lipid"
other-chemicalthing	"sulfur", "uranium", "carbon dioxide"
other-currency	"Travancore Rupee", "$", "lac crore"
other-disease	"bladder cancer", "hypothyroidism", "French Dysentery Epidemic of 1779"
other-educationaldegree	"Master", "Bachelor", "BSc ( Hons ) in physics"
other-god	"Fujin", "Raijin", "El"
other-language	"Latin", "English", "Breton-speaking"
other-law	"Thirty Years ' Peace", "United States Freedom Support Act", "Leahy–Smith America Invents Act ( AIA"
other-livingthing	"monkeys", "insects", "patchouli"
other-medical	"Pediatrics", "amitriptyline", "pediatrician"
person-actor	"Edmund Payne", "Ellaline Terriss", "Tchéky Karyo"
person-artist/author	"George Axelrod", "Hicks", "Gaetano Donizett"
person-athlete	"Tozawa", "Neville", "Jaguar"
person-director	"Richard Quine", "Frank Darabont", "Bob Swaim"
person-other	"Richard Benson", "Campbell", "Holden"
person-politician	"Rivière", "William", "Emeric"
person-scholar	"Wurdack", "Stedman", "Stalmine"
person-soldier	"Joachim Ziegler", "Krukenberg", "Helmuth Weidling"
product-airplane	"Luton", "Spey-equipped FGR.2s", "EC135T2 CPDS"
product-car	"Corvettes - GT1 C6R", "Phantom", "100EX"
product-food	"V. labrusca", "yakiniku", "red grape"
product-game	"Airforce Delta", "Hardcore RPG", "Splinter Cell"
product-other	"PDP-1", "Fairbottom Bobs", "X11"
product-ship	"HMS `` Chinkara ''", "Congress", "Essex"
product-software	"Apdf", "Wikipedia", "AmiPDF"
product-train	"Royal Scots Grey", "High Speed Trains", "55022"
product-weapon	"AR-15 's", "ZU-23-2M Wróbel", "ZU-23-2MR Wróbel II"

Evaluation

Metrics

Label	Precision	Recall	F1
all	0.7041	0.6973	0.7007
art-broadcastprogram	0.5863	0.6252	0.6051
art-film	0.7779	0.752	0.7647
art-music	0.8014	0.7570	0.7786
art-other	0.4209	0.3221	0.3649
art-painting	0.5938	0.6667	0.6281
art-writtenart	0.6854	0.6415	0.6628
building-airport	0.8197	0.8242	0.8219
building-hospital	0.7215	0.8187	0.7671
building-hotel	0.7233	0.6906	0.7066
building-library	0.7588	0.7268	0.7424
building-other	0.5842	0.5855	0.5848
building-restaurant	0.5567	0.4871	0.5195
building-sportsfacility	0.6512	0.7690	0.7052
building-theater	0.6994	0.7516	0.7246
event-attack/battle/war/militaryconflict	0.7800	0.7332	0.7559
event-disaster	0.5767	0.5266	0.5505
event-election	0.5106	0.1319	0.2096
event-other	0.4931	0.4145	0.4504
event-protest	0.3711	0.4337	0.4000
event-sportsevent	0.6156	0.6156	0.6156
location-GPE	0.8175	0.8508	0.8338
location-bodiesofwater	0.7297	0.7622	0.7456
location-island	0.7314	0.6703	0.6995
location-mountain	0.7538	0.7283	0.7409
location-other	0.4370	0.3040	0.3585
location-park	0.7063	0.6878	0.6969
location-road/railway/highway/transit	0.7092	0.7259	0.7174
organization-company	0.6911	0.6943	0.6927
organization-education	0.7799	0.7973	0.7885
organization-government/governmentagency	0.5518	0.4474	0.4942
organization-media/newspaper	0.6268	0.6761	0.6505
organization-other	0.5804	0.5341	0.5563
organization-politicalparty	0.6627	0.7306	0.6949
organization-religion	0.5636	0.6265	0.5934
organization-showorganization	0.6023	0.6086	0.6054
organization-sportsleague	0.6594	0.6497	0.6545
organization-sportsteam	0.7341	0.7703	0.7518
other-astronomything	0.7806	0.8289	0.8040
other-award	0.7230	0.6703	0.6957
other-biologything	0.6733	0.6366	0.6544
other-chemicalthing	0.5962	0.5838	0.5899
other-currency	0.7135	0.7822	0.7463
other-disease	0.6260	0.7063	0.6637
other-educationaldegree	0.6	0.6033	0.6016
other-god	0.7051	0.7118	0.7085
other-language	0.6849	0.7968	0.7366
other-law	0.6814	0.6843	0.6829
other-livingthing	0.5959	0.6443	0.6192
other-medical	0.5247	0.4811	0.5020
person-actor	0.8342	0.7960	0.8146
person-artist/author	0.7052	0.7482	0.7261
person-athlete	0.8396	0.8530	0.8462
person-director	0.725	0.7329	0.7289
person-other	0.6866	0.6672	0.6767
person-politician	0.6819	0.6852	0.6835
person-scholar	0.5468	0.4953	0.5198
person-soldier	0.5360	0.5641	0.5497
product-airplane	0.6825	0.6730	0.6777
product-car	0.7205	0.7016	0.7109
product-food	0.6036	0.5394	0.5697
product-game	0.7740	0.6876	0.7282
product-other	0.5250	0.4117	0.4615
product-ship	0.6781	0.6763	0.6772
product-software	0.6701	0.6603	0.6652
product-train	0.5919	0.6051	0.5984
product-weapon	0.6507	0.5433	0.5921

Uses

Direct Use for Inference

python

from span_marker import SpanMarkerModel

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("tomaarsen/span-marker-mbert-base-fewnerd-fine-super")
# Run inference
entities = model.predict("Most of the Steven Seagal movie \"Under Siege \"(co-starring Tommy Lee Jones) was filmed on the, which is docked on Mobile Bay at Battleship Memorial Park and open to the public.")

Downstream Use

You can finetune this model on your own dataset.

Click to expand

python

from span_marker import SpanMarkerModel, Trainer

# Download from the 🤗 Hub
model = SpanMarkerModel.from_pretrained("tomaarsen/span-marker-mbert-base-fewnerd-fine-super")

# Specify a Dataset with "tokens" and "ner_tag" columns
dataset = load_dataset("conll2003") # For example CoNLL2003

# Initialize a Trainer using the pretrained model & dataset
trainer = Trainer(
    model=model,
    train_dataset=dataset["train"],
    eval_dataset=dataset["validation"],
)
trainer.train()
trainer.save_model("tomaarsen/span-marker-mbert-base-fewnerd-fine-super-finetuned")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Sentence length	1	24.4945	267
Entities per sentence	0	2.5832	88

Training Hyperparameters

learning_rate: 5e-05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 3

Training Results

Epoch	Step	Validation Loss	Validation Precision	Validation Recall	Validation F1	Validation Accuracy
0.2972	3000	0.0274	0.6488	0.6457	0.6473	0.9121
0.5944	6000	0.0252	0.6686	0.6545	0.6615	0.9160
0.8915	9000	0.0239	0.6918	0.6547	0.6727	0.9178
1.1887	12000	0.0235	0.6962	0.6727	0.6842	0.9210
1.4859	15000	0.0233	0.6872	0.6742	0.6806	0.9201
1.7831	18000	0.0226	0.6969	0.6891	0.6929	0.9236
2.0802	21000	0.0231	0.7030	0.6916	0.6973	0.9246
2.3774	24000	0.0227	0.7020	0.6936	0.6978	0.9248
2.6746	27000	0.0223	0.7079	0.6989	0.7034	0.9258
2.9718	30000	0.0222	0.7089	0.7009	0.7049	0.9263

Environmental Impact

Carbon emissions were measured using CodeCarbon.

Carbon Emitted: 0.573 kg of CO2
Hours Used: 3.867 hours

Training Hardware

On Cloud: No
GPU Model: 1 x NVIDIA GeForce RTX 3090
CPU Model: 13th Gen Intel(R) Core(TM) i7-13700K
RAM Size: 31.78 GB

Framework Versions

Python: 3.9.16
SpanMarker: 1.4.1.dev
Transformers: 4.30.0
PyTorch: 2.0.1+cu118
Datasets: 2.14.0
Tokenizers: 0.13.2

Citation

BibTeX

@software{Aarsen_SpanMarker,
    author = {Aarsen, Tom},
    license = {Apache-2.0},
    title = {{SpanMarker for Named Entity Recognition}},
    url = {https://github.com/tomaarsen/SpanMarkerNER}
}