SummaFi - Financial News Summarization System

Project Overview

SummaFi is an advanced financial news summarization system built for my BSc final project. It leverages Facebook's BART model fine-tuned on the CNN/DailyMail dataset to provide accurate, concise summaries of financial news articles. The summarizer achieves ROUGE scores:

• Rouge1: 0.4223
• Rouge2: 0.1935
• RougeL: 0.2889

Core Features

• AI-powered text summarization using fine-tuned BART model
• Automatic article extraction from URLs using newspaper3k
• Financial sentiment analysis using FinBERT
• GDPR-compliant processing (no data storage)
• Interactive web interface using Gradio
• Comprehensive evaluation using ROUGE metrics

Technical Stack

• Base Model: facebook/bart-base
• Dataset: CNN/DailyMail (3.0.0)
• Sentiment Analysis: ProsusAI/finbert
• Framework: PyTorch, HuggingFace Transformers
• Web Interface: Gradio
• Article Extraction: newspaper3k
• Experiment Tracking: Weights & Biases
• Testing: pytest

Installation

Prerequisites

• Python 3.8+
• CUDA-capable GPU (recommended)
• 16GB RAM (recommended)
• Git

Setup Steps

1. Clone repository:

git clone https://github.com/formater/SummaFi.git
cd SummaFi

2. Create virtual environment:

python -m venv .
source bin/activate  # Windows: Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure Weights & Biases:

wandb login

Usage

Training

Fine-tune the model:

python main.py --mode train --config config/config.yaml

Evaluation

Evaluate model performance:

python main.py --mode evaluate --config config/config.yaml --model-path outputs/final_model

Web Interface

Launch the Gradio interface:

python main.py --mode serve --model-path outputs/final_model --port 7860

Project Structure

summa_fi/
├── config/
│   └── config.yaml          # Configuration settings
├── src/
│   ├── data/               # Data processing
│   ├── models/             # Model architecture
│   ├── training/           # Training logic
│   ├── evaluation/         # Evaluation metrics
│   ├── utils/              # Utilities
│   └── web/               # Web interface
├── tests/                  # Test suite
├── docs/                   # Documentation
├── requirements.txt        # Dependencies
└── README.md              # This file

Configuration

Key configuration sections in config/config.yaml:

Model Settings

model:
  name: "facebook/bart-base"
  max_length: 1024
  min_length: 56
  length_penalty: 2.0

Training Settings

training:
  batch_size: 8
  learning_rate: 3e-5
  num_epochs: 3
  warmup_steps: 500

Privacy & Legal Compliance

GDPR Compliance

• No personal data storage
• Real-time processing only
• No cookies or tracking
• Transparent data handling

Copyright Considerations

• Fair use implementation
• No article storage
• Source attribution
• robots.txt compliance

Development Guidelines

Code Style

• PEP 8 compliance
• Type hints (Python 3.8+)
• Comprehensive docstrings
• Clean code principles

Testing

Run tests:

pytest tests/
pytest --cov=src tests/  # with coverage

Troubleshooting

Common Issues

1. Memory Issues

   • Reduce batch size
   • Enable gradient accumulation
   • Use mixed precision training

2. URL Access

   • Verify URL format
   • Check site accessibility
   • Confirm robots.txt compliance

Contributing

1. Fork repository
2. Create feature branch
3. Implement changes
4. Submit pull request

License

MIT License - See LICENSE file

Acknowledgments

• Facebook AI Research (BART)
• HuggingFace Team
• CNN/DailyMail Dataset
• Gradio Team

Contact

• Author: Dudás József
• Email: [email protected]
• GitHub: formater

Testing

Running Tests

# Run all tests
pytest

# Run specific test files
pytest tests/test_data_loader.py
pytest tests/test_summarizer.py
pytest tests/test_article_extractor.py

# Run tests by marker
pytest -m gpu          # Run GPU-specific tests
pytest -m "not slow"   # Skip slow tests
pytest -m integration  # Run integration tests

# Run with coverage report
pytest --cov=src --cov-report=html

Test Categories

• Unit Tests: Individual component testing

  • Data loading and preprocessing
  • Model initialization and inference
  • Article extraction and URL validation

• Integration Tests: Component interaction testing

  • Complete training pipeline
  • End-to-end summarization process
  • Web interface functionality

• GPU Tests: Hardware-specific testing

  • Model GPU utilization
  • Mixed precision training
  • Memory optimization

Coverage Requirements

• Minimum overall coverage: 80%
• Critical paths coverage: 90%
• Web interface coverage: 85%

Continuous Integration

Tests are automatically run on:

• Every push to main branch
• Pull request creation
• Release tag creation