Emotion Classifier — DistilBERT fine-tuned on GoEmotions (8 classes)
Fine-tuned for the Privacy-Preserving Emotional AI Agent project (B.Tech Software Engineering, Delhi Technological University).
Model Details
- Base model: distilbert-base-uncased
- Task: Multi-label emotion classification
- Dataset: GoEmotions (43,410 training samples after preprocessing)
- Classes: joy, sadness, anger, fear, surprise, trust, disgust, neutral
Performance
| Metric | Score |
|---|---|
| Accuracy | 87.2% |
| Precision | 86.1% |
| Recall | 84.9% |
| Macro F1 | 85.5% |
| Latency CPU | ~148ms |
Label Schema
27 GoEmotions fine-grained labels consolidated into 8 coarse categories based on Plutchik's wheel of emotions.
Usage
from src.emotion.detector import EmotionDetector
detector = EmotionDetector()
result = detector.detect("I feel really happy today!")
print(result.primary_emotion) # "joy"
- Downloads last month
- 6