README.md

metadata

language:
  - en
license: apache-2.0
library_name: transformers
tags:
  - emotion-classification
  - healthcare
  - distilbert
  - patient-doctor-conversations
  - text-classification
  - clinical-AI
  - mental-health
model_index:
  - name: patient-emotion-classifier
    results:
      - task:
          type: text-classification
        metrics:
          - type: accuracy
            value: 0.713
          - type: f1
            value: 0.722

🤖 Patient Emotion Classifier

Advanced AI-Powered Emotion Recognition for Healthcare Dialogues

Part of the Blended AI+X Initiative — Bridging Artificial Intelligence and Healthcare

---

🔬 Overview

We are thrilled to introduce Patient Emotion Classifier, a state-of-the-art NLP model engineered to understand emotional nuances in patient-doctor conversations.

This model represents our commitment to advancing AI for Healthcare (AI+X), leveraging cutting-edge transformer architectures to bridge the gap between artificial intelligence and compassionate care.

Key Capabilities

• Multiclass Emotion Recognition — Identifies 6 distinct emotional states in clinical dialogues
• Healthcare-Optimized — Specifically trained on medical conversation data
• Production-Ready — Deployable via REST API for real-time inference
• Lightweight & Efficient — Built on DistilBERT for fast inference

🎯 Emotion Categories

Our model classifies emotional states into 6 clinically-relevant categories:

Category	Description
😐 Neutral	Objective, non-emotional statements
😰 Anxiety/Fear	Patient expresses worry, concern, or fear
😠 Anger/Frustration	Patient shows frustration or displeasure
😢 Sadness/Helplessness	Patient feels down or hopeless
🤔 Confusion/Doubt	Patient expresses uncertainty or questions
🙏 Gratitude/Relief	Patient conveys thanks or relief

📊 Model Performance

Overall Metrics

Metric	Value
Accuracy	71.3%
Macro F1	0.722
Weighted F1	0.72

Per-Class Performance

Emotion	Precision	Recall	F1-Score
Neutral	0.75	0.78	0.76
Anxiety/Fear	0.52	0.63	0.57
Anger/Frustration	0.80	0.73	0.76
Sadness/Helplessness	0.65	0.55	0.60
Confusion/Doubt	0.60	0.58	0.59
Gratitude/Relief	0.72	0.75	0.73

Label Distribution

🚀 Quick Start

1. Install Dependencies

pip install -r requirements.txt

### 2. Launch the Service

```bash
cd see
python app.py

### 3. Access the Interface

http://localhost:8002

## 📚 Dataset

This model was trained on a meticulously curated subset of medical dialogues:

- **Original Source**: [Chinese MedDialog Dataset](https://tianchi.aliyun.com/dataset/92110) — Alibaba Cloud Tianchi
- **Post-Processing**: Carefully filtered, translated, and annotated for emotion classification
- **Total Samples**: 28,280 annotated dialogues
- **Categories**: 6 emotion labels
- **Language**: English

## 📚 References

1. **MedDialog Dataset**  
   Chinese Medical Dialogue Dataset. Alibaba Cloud Tianchi.  
   https://tianchi.aliyun.com/dataset/92110

2. **DistilBERT**  
   Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019). *DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter*. arXiv preprint arXiv:1910.01108.  
   https://arxiv.org/abs/1910.01108

## 🏗️ Technical Specifications

| Component | Details |
|-----------|---------|
| **Base Architecture** | [DistilBERT](https://huggingface.co/distilbert/distilbert-base-uncased) |
| **Task Type** | 6-class emotion classification |
| **Max Sequence Length** | 512 tokens |
| **Framework** | PyTorch + Transformers |

## 📁 Project Structure

patient-emotion-analysis/
├── best_model/          # Fine-tuned model weights
├── see/                 # Inference service
│   ├── app.py          # Web application
│   ├── inference.py    # Core inference logic
│   └── templates/      # UI templates
├── data/               # Training & evaluation data
├── requirements.txt    # Dependencies
└── README.md           # This file

---

<div align="center">

**Blended AI+X Initiative** — *Advancing Healthcare Through Intelligence*

</div>