bert-ner-turkish-cased

This model is a fine-tuned version of dbmdz/bert-base-turkish-cased on a custom Turkish NER dataset. It achieves the following results on the evaluation set:

• Loss: 0.0987
• Precision: 0.9112
• Recall: 0.9364
• F1: 0.9236
• Accuracy: 0.9600

Model description

This model identifies named entities in Turkish text:

LABELS = [
    "O", "B-PER", "I-PER", "B-LOC", "I-LOC", "B-ORG", "I-ORG",
    "B-DATE", "I-DATE", "B-MONEY", "I-MONEY", "B-MISC", "I-MISC"
]

• PER: Person
• LOC: Location
• ORG: Organization
• DATE: Date
• MONEY: Money
• MISC: Miscellaneous Entities

Intended uses & limitations

Extracting entities from Turkish text in NLP pipelines.

How to Use

from transformers import pipeline

model_name = "yeniguno/bert-ner-turkish-cased"

ner_pipeline = pipeline("ner", model=model_name, tokenizer=model_name, aggregation_strategy="simple")

text = """Selim Parlak, 2023-11-15 tarihinde, CUMHURİYET MAH. DUMAN SOKAK 22500 HAVSA/EDİRNE adresinden, Dünya Varlık Yönetim A.Ş. aracılığıyla 850 TRY değerindeki MP.2386.JPA.IP5.WHT.I İPHONE5 ŞARJLI KILIF "AİR" 1700 MAH (BEYAZ) ürününü satın aldı."""

results = ner_pipeline(text)

for result in results:
    print(result)

"""
{'entity_group': 'PER', 'score': 0.9993254, 'word': 'Selim Parlak', 'start': 0, 'end': 12}
{'entity_group': 'DATE', 'score': 0.9987677, 'word': '2023 - 11 - 15', 'start': 14, 'end': 24}
{'entity_group': 'LOC', 'score': 0.99951524, 'word': 'CUMHURİYET MAH. DUMAN SOKAK 22500 HAVSA / EDİRNE', 'start': 36, 'end': 82}
{'entity_group': 'ORG', 'score': 0.8487069, 'word': 'Dünya Varlık Yönetim A. Ş.', 'start': 95, 'end': 120}
{'entity_group': 'MONEY', 'score': 0.9970985, 'word': '850 TRY', 'start': 134, 'end': 141}
{'entity_group': 'MISC', 'score': 0.97721404, 'word': 'MP. 2386. JPA. IP5. WHT. I İPHONE5 ŞARJLI KILIF " AİR " 1700 MAH ( BEYAZ )', 'start': 154, 'end': 219}
"""

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
0.1351	1.0	1527	0.1158	0.8592	0.9070	0.8825	0.9517
0.1088	2.0	3054	0.1045	0.8787	0.9336	0.9053	0.9574
0.1016	3.0	4581	0.0993	0.8901	0.9280	0.9086	0.9576
0.1102	4.0	6108	0.0963	0.8991	0.9277	0.9132	0.9587
0.0877	5.0	7635	0.0953	0.9046	0.9292	0.9167	0.9584
0.0933	6.0	9162	0.0939	0.9036	0.9321	0.9176	0.9593
0.0827	7.0	10689	0.0967	0.8986	0.9398	0.9188	0.9605
0.0933	8.0	12216	0.0949	0.9122	0.9292	0.9206	0.9593
0.084	9.0	13743	0.0987	0.9112	0.9364	0.9236	0.9600

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0