DataWizardd's picture
Update README.md
d578662 verified
---
language: ko
datasets:
- investing_comments_krw
metrics:
- accuracy
- precision
- recall
- f1
tags:
- sentiment-analysis
- finance
- krw
- binary-classification
- emotion
model-index:
- name: FinBERT-Sentiment-KRW-Comment (v3)
results:
- task:
type: text-classification
name: Sentiment Analysis (Fear vs. Greed)
metrics:
- type: accuracy
value: 0.94
- type: precision
value: 0.94
- type: recall
value: 0.94
- type: f1
value: 0.94
---
# FinBERT-Sentiment-KRW-Comment (v3)
이 λͺ¨λΈμ€ [`snunlp/KR-FinBERT-SC`](https://huggingface.co/snunlp/KR-FinBERT-SC)λ₯Ό 기반으둜 νŒŒμΈνŠœλ‹ν•œ ν•œκ΅­μ–΄ 금육 감정 뢄석 λͺ¨λΈμž…λ‹ˆλ‹€.
특히 **ν™˜μœ¨(FX) κ΄€λ ¨ λŒ“κΈ€**μ—μ„œ λ‚˜νƒ€λ‚˜λŠ” 감정을 **곡포(0)** λ˜λŠ” **μš•μ‹¬(1)** 이진 λΆ„λ₯˜ν•˜λŠ” 데 λͺ©μ μ΄ μžˆμŠ΅λ‹ˆλ‹€.
## 🧾 라벨 μ •μ˜
| 라벨 | μ„€λͺ… |
|------|------------|
| 0 | 곡포 (Fear) |
| 1 | μš•μ‹¬ (Greed) |
## πŸ‹οΈβ€β™‚οΈ ν•™μŠ΅ 정보
- **Base model**: `snunlp/KR-FinBERT-SC`
- **Task**: 감정 이진 λΆ„λ₯˜ (곡포 vs μš•μ‹¬)
- **Input**: ν•œκ΅­μ–΄ λŒ“κΈ€ (`content`)
- **Output**: 0 λ˜λŠ” 1
- **Training epochs**: 4
- **Train size**: μ•½ `X` 개
- **Eval size**: μ•½ `X` 개
- **Evaluation metric**: Accuracy, Precision, Recall, F1
### πŸ“Š 평가 κ²°κ³Ό (Test Set κΈ°μ€€)
| Metric | Score |
|------------|-------|
| Accuracy | 0.94 |
| Precision | 0.94 |
| Recall | 0.94 |
| F1-score | 0.94 |
> 곡포(0): precision=0.95, recall=0.91, f1=0.93
> μš•μ‹¬(1): precision=0.93, recall=0.96, f1=0.94
## πŸ§ͺ μ‚¬μš© μ˜ˆμ‹œ
```python
from transformers import pipeline
pipe = pipeline("text-classification", model="DataWizardd/finbert-sentiment-krw-comment-v3")
pipe("μ‹¬μƒμΉ˜ μ•Šλ„€μš” μ€‘λ™μ „μŸ λ°œλ°œν•˜κ³  1μ›”8일에 김정은이 미사일 μ‹€ν—˜ν•˜κ³  그러면 1170은 기본이고 1190원도 μˆœμ‹κ°„μΌλ“―.")
# β†’ [{'label': '1', 'score': 0.98}]