metadata
language: ko
datasets: naver-finance-news
tags:
- sentiment-analysis
- korean
- finance
- finbert
- transformers
license: mit
model-index:
- name: finbert-sentiment-ko
results:
- task:
name: Sentiment Analysis
type: text-classification
metrics:
- type: accuracy
value: 0.93
FinBERT Sentiment Analysis (Korean, Finance Domain)
μ΄ λͺ¨λΈμ νκ΅μ΄ νμ¨(κΈμ΅) λ΄μ€ μμ½λ¬Έμ λμμΌλ‘ κ°μ μ λΆλ₯νκΈ° μν΄ νμΈνλλ BERT κΈ°λ° λͺ¨λΈμ
λλ€.
κ°μ λΆλ₯λ λ€μ μΈ κ°μ§ ν΄λμ€ μ€ νλλ‘ μνλ©λλ€:
0
: λΆμ 1
: μ€λ¦½2
: κΈμ
π§ νμ΅ μ 보
- κΈ°λ° λͺ¨λΈ:
snunlp/KR-FinBERT-SC
- λ°μ΄ν°: μ§μ μμ§ν λ€μ΄λ² νμ¨(κΈμ΅) λ΄μ€ μμ½ + κ°μ μμμ λΌλ²¨λ§
- μ΄ μν μ: μ½ 200
- Optimizer: AdamW
- Epochs: 4
- μ΅λ κΈΈμ΄: 128
- νκ° μ§ν: Accuracy, F1 Score
π μ±λ₯ νκ°
κ°μ ν΄λμ€ | Precision | Recall | F1-score | Support |
---|---|---|---|---|
λΆμ | 0.89 | 1.00 | 0.94 | 17 |
μ€λ¦½ | 1.00 | 0.82 | 0.90 | 11 |
κΈμ | 0.93 | 0.93 | 0.93 | 14 |
μ νλ | 0.93 | 42 |
μ 체 μ νλ: 93%
Macro F1-score: 0.92
π μ¬μ© λ°©λ²
from transformers import pipeline
pipe = pipeline("text-classification", model="DataWizardd/finbert-sentiment-ko")
pipe("νμ¨μ΄ κΈλ±νλ©° μμ₯ λΆμμ΄ μ»€μ§κ³ μλ€.")
# μΆλ ₯: [{'label': 'λΆμ ', 'score': 0.95}]