bayrameker
commited on
Update README.md
Browse files
README.md
CHANGED
@@ -8,4 +8,110 @@ metrics:
|
|
8 |
- accuracy
|
9 |
base_model:
|
10 |
- answerdotai/ModernBERT-base
|
11 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
- accuracy
|
9 |
base_model:
|
10 |
- answerdotai/ModernBERT-base
|
11 |
+
---
|
12 |
+
|
13 |
+
```markdown
|
14 |
+
# Turkish Sentiment Modern BERT
|
15 |
+
```
|
16 |
+
This model is a fine-tuned **ModernBERT** for **Turkish Sentiment Analysis**. It was trained on the [winvoker/turkish-sentiment-analysis-dataset](https://huggingface.co/datasets/winvoker/turkish-sentiment-analysis-dataset) and is designed to classify Turkish text into sentiment categories, such as **Positive**, **Negative**, and **Neutral**.
|
17 |
+
|
18 |
+
## Model Overview
|
19 |
+
|
20 |
+
- **Model Type**: ModernBERT (BERT variant)
|
21 |
+
- **Task**: Sentiment Analysis
|
22 |
+
- **Languages**: Turkish
|
23 |
+
- **Dataset**: [winvoker/turkish-sentiment-analysis-dataset](https://huggingface.co/datasets/winvoker/turkish-sentiment-analysis-dataset)
|
24 |
+
- **Labels**: Positive, Negative, Neutral
|
25 |
+
- **Fine-Tuning**: Fine-tuned for sentiment classification.
|
26 |
+
|
27 |
+
## Performance Metrics
|
28 |
+
|
29 |
+
The model was trained for **2 epochs** with the following results:
|
30 |
+
|
31 |
+
| Epoch | Training Loss | Validation Loss | Accuracy | F1 Score |
|
32 |
+
|-------|---------------|-----------------|-----------|-----------|
|
33 |
+
| 1 | 0.2182 | 0.1920 | 92.16% | 84.57% |
|
34 |
+
| 2 | 0.1839 | 0.1826 | 92.58% | 86.05% |
|
35 |
+
|
36 |
+
- **Training Loss**: Measures the model's fit to the training data.
|
37 |
+
- **Validation Loss**: Measures the model's generalization to unseen data.
|
38 |
+
- **Accuracy**: The percentage of correct predictions over all examples.
|
39 |
+
- **F1 Score**: A balanced metric between precision and recall.
|
40 |
+
|
41 |
+
## Model Inference Example
|
42 |
+
|
43 |
+
Here’s an example of how to use the model for sentiment analysis of Turkish text:
|
44 |
+
|
45 |
+
```python
|
46 |
+
from transformers import AutoModelForSequenceClassification, AutoTokenizer
|
47 |
+
import torch
|
48 |
+
|
49 |
+
# Load the pre-trained model and tokenizer
|
50 |
+
model_name = "your_huggingface_username/turkish-sentiment-modern-bert"
|
51 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
52 |
+
model = AutoModelForSequenceClassification.from_pretrained(model_name)
|
53 |
+
|
54 |
+
# Example texts for prediction
|
55 |
+
texts = ["bu ürün çok iyi", "bu ürün berbat"]
|
56 |
+
|
57 |
+
# Tokenize the inputs
|
58 |
+
inputs = tokenizer(texts, padding=True, truncation=True, return_tensors="pt")
|
59 |
+
|
60 |
+
# Make predictions
|
61 |
+
with torch.no_grad():
|
62 |
+
logits = model(**inputs).logits
|
63 |
+
|
64 |
+
# Get the predicted sentiment labels
|
65 |
+
predictions = torch.argmax(logits, dim=-1)
|
66 |
+
labels = ["Negative", "Neutral", "Positive"] # Adjust based on your label mapping
|
67 |
+
for text, pred in zip(texts, predictions):
|
68 |
+
print(f"Text: {text} -> Sentiment: {labels[pred.item()]}")
|
69 |
+
```
|
70 |
+
|
71 |
+
### Example Output:
|
72 |
+
|
73 |
+
```
|
74 |
+
Text: bu ürün çok iyi -> Sentiment: Positive
|
75 |
+
Text: bu ürün berbat -> Sentiment: Negative
|
76 |
+
```
|
77 |
+
|
78 |
+
## Installation
|
79 |
+
|
80 |
+
To use this model, first install the required dependencies:
|
81 |
+
|
82 |
+
```bash
|
83 |
+
pip install transformers
|
84 |
+
pip install torch
|
85 |
+
pip install datasets
|
86 |
+
```
|
87 |
+
|
88 |
+
## Model Card
|
89 |
+
|
90 |
+
- **Model Name**: turkish-sentiment-modern-bert
|
91 |
+
- **Hugging Face Repo**: [Link to Model Repository](https://huggingface.co/your_huggingface_username/turkish-sentiment-modern-bert)
|
92 |
+
- **License**: MIT (or another applicable license)
|
93 |
+
- **Author**: Bayram Eker
|
94 |
+
- **Date**: 2024-12-21
|
95 |
+
|
96 |
+
## Training Details
|
97 |
+
|
98 |
+
- **Model**: ModernBERT (Base variant)
|
99 |
+
- **Framework**: PyTorch
|
100 |
+
- **Training Time**: 34 minutes (2 epochs)
|
101 |
+
- **Batch Size**: 32
|
102 |
+
- **Learning Rate**: 8e-5
|
103 |
+
- **Optimizer**: AdamW
|
104 |
+
|
105 |
+
## Acknowledgments
|
106 |
+
|
107 |
+
- The model was trained on the [winvoker/turkish-sentiment-analysis-dataset](https://huggingface.co/datasets/winvoker/turkish-sentiment-analysis-dataset).
|
108 |
+
- Special thanks to the Hugging Face community and all contributors to the transformers library.
|
109 |
+
|
110 |
+
## Future Work
|
111 |
+
|
112 |
+
- Expand the model with more complex sentiment labels (e.g., multi-class sentiment, aspect-based sentiment analysis).
|
113 |
+
- Fine-tune the model on a larger, more diverse dataset for better generalization across various domains.
|
114 |
+
|
115 |
+
## License
|
116 |
+
|
117 |
+
This model is licensed under the MIT License. See the LICENSE file for more details.
|