About the Model

This model is a Turkish-based convBERT model created to classify Turkish social media bullying data. Classes included in the model

Cinsiyetçilik
Irkçılık
Kızdırma
Nötr

The model was trained from a dataset containing 2981 Twitter data.

The success rates achieved in training the model are as follows:

	Cinsiyetçilik	Irkçılık	Kızdırma	Nötr
Precision	0.9	0.89	0.85	0.90
Recall	0.9	0.91	0.88	0.86
F1 Score	0.9	0.90	0.86	0.88

Accuracy: 0.88

Confusion matrix of the model:

Dependency

pip install transformers

pip install torch

Example

from transformers import ConvBertTokenizer, ConvBertForSequenceClassification,TextClassificationPipeline

tokenizer = ConvBertTokenizer.from_pretrained("AIZinu/convBERT-turk-based-cyberbullying")
model = ConvBertForSequenceClassification.from_pretrained("AIZinu/convBERT-turk-based-cyberbullying")

pipe= TextClassificationPipeline(model=model, tokenizer=tokenizer)

print(pipe("Hadi modeli deneyelim"))

Result:

[{'label': 'LABEL_3', 'score': 0.9994651675224304}]

Model Card Authors

Bilge Nur BEKAR