README.md

---
title: Financial Sentiment Analysis Ensemble
emoji: 📈
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
license: apache-2.0
tags:
- financial-sentiment
- sentiment-analysis
- finance
- nlp
- ensemble
- transformers
- gradio
---

# 🚀 Financial Sentiment Analysis Ensemble

An advanced AI-powered sentiment analysis system specifically designed for financial texts. This application uses an ensemble of three fine-tuned transformer models to provide highly accurate sentiment predictions for financial news, social media posts, and market commentary.

## 🎯 Features

- **Ensemble Prediction**: Combines predictions from 3 specialized models for higher accuracy
- **Real-time Analysis**: Instant sentiment analysis with confidence scores
- **Interactive Interface**: User-friendly Gradio interface with examples
- **Detailed Results**: Individual model predictions and probability distributions
- **Financial Focus**: Specifically trained on financial datasets

## 🧠 Model Architecture

The ensemble consists of three fine-tuned models:

1. **DistilBERT Model** (`codealchemist01/financial-sentiment-distilbert`)
   - Fast and efficient for real-time analysis
   - Based on DistilBERT-base-uncased
   - Optimized for speed without sacrificing accuracy

2. **BERT-Large Model** (`codealchemist01/financial-sentiment-bert-large`)
   - High accuracy with deep contextual understanding
   - Based on BERT-Large-uncased
   - Superior performance on complex financial texts

3. **Improved Model** (`codealchemist01/financial-sentiment-improved`)
   - Enhanced with advanced training techniques
   - Balanced dataset training
   - Custom loss functions and optimization

## 📊 Performance

- **Ensemble Accuracy**: 79.7%
- **Individual Model Accuracies**: 79.7% (DistilBERT), 84.3% (BERT-Large), 82.1% (Improved)
- **Dataset**: Twitter Financial News Sentiment
- **Categories**: Bearish 📉, Neutral ➡️, Bullish 📈

## 🚀 Usage

### Web Interface

Simply enter your financial text in the input box and click "Analyze Sentiment" to get:
- Ensemble prediction with confidence score
- Probability distribution across all sentiment categories
- Individual predictions from each model
- Visual probability chart

### API Usage

You can also use the individual models directly:

```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load any of the models
model_name = "codealchemist01/financial-sentiment-distilbert"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

def predict_sentiment(text):
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)
    
    with torch.no_grad():
        outputs = model(**inputs)
        predictions = torch.nn.functional.softmax(outputs.logits, dim=-1)
    
    labels = ["Bearish", "Neutral", "Bullish"]
    predicted_class = torch.argmax(predictions, dim=-1).item()
    confidence = predictions[0][predicted_class].item()
    
    return {
        "label": labels[predicted_class],
        "confidence": confidence
    }

# Example
result = predict_sentiment("The stock market is showing strong growth today")
print(result)
```

## 📈 Example Predictions

- **"Tesla's innovative battery technology could revolutionize the automotive industry."**
  - Prediction: Bullish 📈 (85.2% confidence)

- **"Company earnings fell short of expectations, leading to a significant drop in share price."**
  - Prediction: Bearish 📉 (91.7% confidence)

- **"The Federal Reserve maintained interest rates, keeping market conditions stable."**
  - Prediction: Neutral ➡️ (78.3% confidence)

## 🛠️ Technical Details

- **Framework**: Transformers, PyTorch
- **Interface**: Gradio 4.0+
- **Training**: Fine-tuned on financial datasets with advanced techniques
- **Ensemble Method**: Average probability aggregation
- **Preprocessing**: Text normalization and tokenization

## 📝 License

This project is licensed under the Apache 2.0 License.

## 🤝 Contributing

Contributions are welcome! Please feel free to submit issues or pull requests.

## 📧 Contact

For questions or collaborations, please reach out through the Hugging Face community.

---

*Built with ❤️ for the financial AI community*