finbert-sentiment-tr
A fine-tuned version of distilbert/distilbert-base-uncased for financial
text sentiment classification (negative / neutral / positive).
Usage
```python from transformers import pipeline
classifier = pipeline("text-classification", model="leplebi/finbert-sentiment-tr", top_k=None) classifier("The company reported strong quarterly earnings.") ```
Training Details
- Base model: distilbert/distilbert-base-uncased
- Dataset: Financial PhraseBank (Malo et al., 2014), "50Agree" configuration (~4846 sentences)
- Epochs: 4 (best checkpoint selected at epoch 3, based on
f1_macro) - Learning rate: 2e-5
- Batch size: 16
Results (test set, 727 examples)
| Metric | Value |
|---|---|
| Accuracy | 0.8638 |
| F1 Macro | 0.8497 |
| F1 Weighted | 0.8636 |
| Precision Macro | 0.8372 |
| Recall Macro | 0.8675 |
Limitations
- Trained on short, single-sentence financial statements; performance on longer reports or paragraphs may differ.
- Not intended for use on non-English text.
License / Data Source Note
This model was trained using the Financial PhraseBank dataset (Malo et al., 2014), which is distributed under an academic/research license. Please review the original license terms before any commercial use.