README.md

---
title: Emotion Detection
emoji: 🎭
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: mit
---

# 🎭 Multi-Label Emotion Detection

Detect multiple emotions in text using **DeBERTa-v3-base** fine-tuned for multi-label emotion classification.

## Features

- **Multi-Label Classification**: Detect multiple emotions simultaneously
- **5 Emotions**: Anger, Fear, Joy, Sadness, Surprise
- **State-of-the-Art Model**: DeBERTa-v3-base (184M parameters)
- **High Performance**: F1 Score ~0.85 on validation set

## Model Architecture

- **Base Model**: microsoft/deberta-v3-base
- **Task**: Multi-label emotion classification
- **Training**: Fine-tuned on emotion detection dataset
- **Parameters**: 184 million

## How to Use

1. Enter or paste text in the input box
2. Click "Analyze Emotions" or press Enter
3. View detected emotions and their probabilities
4. Try the example texts for demonstration

## Performance

- **Validation F1 Score**: ~0.85
- **Validation Accuracy**: ~0.55
- **Training**: 12 epochs with early stopping
- **Optimizer**: AdamW with linear warmup

## Dataset

Trained on multi-label emotion classification dataset with:
- 6,827 training samples
- 1,707 test samples
- 5 emotion labels

## Examples

**Happy**:
> "I just got accepted into my dream university! I can't believe it!"

**Fearful**:
> "I'm so worried about the exam tomorrow. I haven't studied enough."

**Mixed Emotions**:
> "I'm frustrated with this project but also excited about the possibilities."

## Technical Details

- **Framework**: PyTorch, Transformers, Gradio
- **Tokenizer**: DeBERTa-v3-base tokenizer
- **Max Length**: 160 tokens
- **Threshold**: 0.5 for emotion detection

## License

MIT License

## Acknowledgments

- Model: Microsoft DeBERTa-v3-base
- Framework: Hugging Face Transformers
- UI: Gradio

---

Built with ❤️ for Deep Learning and Gen AI course