bert-base-turkish-sentiment-analysis

This model is a fine-tuned version of dbmdz/bert-base-turkish-cased on an winvoker/turkish-sentiment-analysis-dataset (The shuffle function was used with a training dataset of 10,000 data points and a test dataset of 2,000 points.). It achieves the following results on the evaluation set:

• Loss: 0.2458
• Accuracy: 0.962

Model description

• "Positive" : LABEL_1
• "Notr" : LABEL_0
• "Negative" : LABEL_2
• "Fine-Tuning Process" : https://github.com/saribasmetehan/Transformers-Library/blob/main/Turkish_Text_Classifiaction_Fine_Tuning_PyTorch.ipynb

Example

from transformers import pipeline
text = "senden nefret ediyorum"
model_id = "saribasmetehan/bert-base-turkish-sentiment-analysis"
classifer = pipeline("text-classification",model = model_id)
preds= classifer(text)
print(preds)

#[{'label': 'LABEL_2', 'score': 0.7510055303573608}]

Load model directly

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("saribasmetehan/bert-base-turkish-sentiment-analysis")
model = AutoModelForSequenceClassification.from_pretrained("saribasmetehan/bert-base-turkish-sentiment-analysis")

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• 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
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.1902	1.0	625	0.1629	0.9575
0.1064	2.0	1250	0.1790	0.96
0.0631	3.0	1875	0.2358	0.96
0.0146	4.0	2500	0.2458	0.962

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1