Skip to content

It is a Turkish BERT-based model that will analyze people's bank complaints and classify them according to one of eight categories.

License

Notifications You must be signed in to change notification settings

elifftosunn/Bert-Bank-Model

Repository files navigation

Bert-Bank-Model

By clicking on the icon, you can use the "bert-bank-model" model over the Hugging Face.

About the Bert-Bank-Model

It is a Turkish BERT-based model that will analyze people's bank complaints and classify them according to one of eight categories. The classification headings are as follows:

  • Diğer
  • Hesap İşlemleri
  • KKB Skor
  • Kargo
  • Kart İşlemleri
  • Kredi İşlemleri
  • Limit
  • Müşteri Temsilcisi

label_percent

246412 thousand complaints were used in model training. The success rates in education are as follows.

Kart İşlemleri Kredi İşlemleri Hesap İşlemleri Kargo Limit Müşteri Temsilcisi KKB Skor accuracy
Precision 0.977292 0.971119 0.985294 0.953096 0.98616 0.989115 0.991824 0.982336
Recall 0.978114 0.960714 0.985294 0.986348 0.978590 0.982224 0.992679 0.982336
F1 Score 0.977703 0.965889 0.985294 0.969437 0.983577 0.985657 0.992251 0.982336

Example

from transformers import AutoTokenizer, TextClassificationPipeline, TFBertForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("elifftosunn/Bert-Bank-Model")
model = TFBertForSequenceClassification.from_pretrained("elifftosunn/Bert-Bank-Model", from_pt=True)
pipe = TextClassificationPipeline(model=model, tokenizer=tokenizer)

print(pipe('QNB Finansbank 1.39 oranlı 50.000 TL yeni müşterilere özel ihtiyaç kredisi 1.92 oranında veriyor amaç hesap açtırmak kampanyanın hiçbir gerçekçiliği yoktur. Resmen milletle dalga geçiyorsunuz. Ne demek oluyor bu. 1,39 dan kredi deyip içeriğine girince 2 katına çıkıyor. Böyle saçma bir banka'))

Result

[{'label': 'Kredi İşlemleri', 'score': 0.9589990377426147}]