NewsSentimentScanner 📊

A complete sentiment analysis toolkit combining Google News RSS aggregation with FinBERT (finance-tuned transformer) for real-time market sentiment tracking.

Features

1. Notebook Pipeline (test.ipynb)

• Fetch latest 10 Google News items per query
• Multi-locale support (US, GB, CA, IN) with query variants
• Per-article sentiment analysis with FinBERT or VADER
• Aggregated summary across all queries and articles
• Focused on INTC/Intel stock queries

2. Web App (app.py)

• Flask-based REST API
• Modern, responsive web interface
• Real-time FinBERT sentiment analysis
• Displays:
  • Primary sentiment label (Positive/Neutral/Negative)
  • Probability scores with visual progress bars
  • Compound score for quick reference
• Handles long texts with automatic truncation (max 512 tokens)

Installation

1. Clone and setup:

git clone https://github.com/RachGranville/sentimentstrading.git
cd sentimentstrading
python -m venv .venv
.venv\Scripts\activate  # On Windows
source .venv/bin/activate  # On macOS/Linux

2. Install dependencies:

pip install -r requirements.txt

Usage

Option A: Jupyter Notebook (Batch Analysis)

jupyter notebook test.ipynb

• Edit queries list (line ~12 in the code cell)
• Choose method = "finbert" or "vader"
• Run the Google News cell to analyze 100+ articles (10 per query)
• View aggregated sentiment summary

Option B: Flask Web App (Interactive)

python app.py

Then open http://127.0.0.1:5000 in your browser and:

• Type or paste text
• Click "Analyze Sentiment" (or press Enter)
• See instant FinBERT results

Project Structure

NewsSentimentScanner/
├── app.py                 # Flask web app
├── test.ipynb            # Jupyter notebook for batch analysis
├── sentiment_analysis.py  # Utility functions
├── requirements.txt      # Python dependencies
├── templates/
│   └── index.html        # Web UI
└── README.md

Key Technologies

• FinBERT: Financial sentiment analysis (yiyanghkust/finbert-tone)
• VADER: Alternative sentiment analyzer (nltk)
• feedparser: Google News RSS parsing
• Flask: Web framework
• Transformers: HuggingFace model pipeline

Sentiment Scores Explained

FinBERT Output

• Positive: Probability of positive sentiment (0.0–1.0)
• Neutral: Probability of neutral sentiment (0.0–1.0)
• Negative: Probability of negative sentiment (0.0–1.0)
• Compound: Pseudo-compound score = Positive − Negative (−1.0 to 1.0)

Classification

• Positive: Highest probability is Positive class
• Neutral: Highest probability is Neutral class
• Negative: Highest probability is Negative class

Notes

• First FinBERT run downloads the model (~500 MB)
• Long texts automatically truncated to 512 tokens (BERT limit)
• Google News RSS queries fetch up to 10 articles with multi-locale fallbacks
• Flask runs in debug mode; use WSGI server for production

License

MIT