Financial News Sentiment Prediction using Deep Learning & FinBERT

A production-ready sentiment classification system that classifies finance-related news and tweets into Bearish 🔴, Bullish 🟢, or Neutral 🔵 using deep learning and transformer-based models.

The project implements three RNN-based baseline models (SimpleRNN, LSTM, and GRU) and fine-tunes FinBERT (ProsusAI/finbert) to achieve state-of-the-art performance on financial sentiment analysis. An interactive Streamlit dashboard is included for real-time sentiment prediction.

Live Demo: https://ntini97-financial-sentiment-dashboard.hf.space/

---

Project Structure

financial-news-sentiment/
│
├── .dockerignore                                               ← Docker ignore rules
├── .gitignore                                                  ← Git ignore rules
├── Dockerfile                                                  ← Docker configuration
├── requirements.txt                                            ← Python dependencies
│
├── app.py                                                      ← Streamlit dashboard
├── export_vocab.py                                             ← Exports tokenizer vocabulary for the dashboard
│
├── Financial_News_Sentiment_Prediction_Complete.ipynb          ← End-to-end notebook (EDA, preprocessing, model training & evaluation)
│                                  
│
├── project_report.md                                           ← Detailed project report
├── README.md                                                   ← Project documentation
│
├── Assets/                                                     ← Images and other static assets
│
├── models/                                                     ← Saved trained models and tokenizer files
│
└── .vscode/                                                    ← VS Code workspace settings (optional)

Note: The models/ directory is not included in this repository because it contains large trained model weights (e.g., a fine-tuned FinBERT model of approximately 500 MB). Run the notebook to train the models and generate the required files locally before launching the Streamlit application.

---

Setup & Installation

1. Clone the repository

git clone https://github.com/nithansantiago021/Financial-News-Sentiment-Prediction-using-Deep-Learning-BERT.git
cd Financial-News-Sentiment-Prediction-using-Deep-Learning-BERT

2. Create a virtual environment (recommended)

python -m venv venv
source venv/bin/activate        # Linux / macOS
venv\Scripts\activate           # Windows

3. Install dependencies

pip install -r requirements.txt

---

How to Run

Step 1 — Train all models (notebook)

Open and run every cell in the notebook from top to bottom:

jupyter notebook Financial_News_Sentiment_Prediction_Complete.ipynb

This will:

• Download the dataset from HuggingFace automatically
• Preprocess tweets with the full cleaning pipeline
• Train SimpleRNN, LSTM, and GRU with early stopping
• Fine-tune FinBERT (GPU recommended; pre-recorded results used on CPU)
• Save model weights to models/

Step 2 — Export vocabulary

python export_vocab.py

This rebuilds the word→index mapping using the exact same pipeline as the notebook and saves it to models/vocab.json. The dashboard needs this file to convert new text to integers at inference time.

# Optional: verify vocab is consistent with saved model weights
python export_vocab.py --verify

Step 3 — Launch the Streamlit dashboard

streamlit run app.py

---