README.md

# 🎮 KcELECTRA Steam Review Aspect Classifier (Phase 1)

Fine-tuned model based on **beomi/KcELECTRA-base**  
for **Aspect-Based Sentiment Analysis (ABSA)** on Steam game reviews.

## 📘 Model Info
- Base model: `beomi/KcELECTRA-base`
- Task: Multi-label classification (6 aspects)
- Labels:
  - STORY
  - OPTIMIZATION
  - GRAPHICS
  - PRICE_VALUE
  - BALANCE
  - ENGAGEMENT

## ⚙️ Training
- Dataset: Custom labeled Steam reviews (2,349 samples)
- Loss: BCEWithLogitsLoss
- Epochs: 5
- LR: 2e-5
- Batch size: 16

## 🧠 Usage Example
```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification

model = AutoModelForSequenceClassification.from_pretrained("Wing4/kcelectra-steam-aspect-classifier")
tokenizer = AutoTokenizer.from_pretrained("Wing4/kcelectra-steam-aspect-classifier")

inputs = tokenizer("그래픽은 좋지만 최적화가 별로야", return_tensors="pt")
outputs = model(**inputs)
print(outputs.logits.sigmoid())  # 각 측면별 확률