README.md

metadata

title: FinBERT Sentiment Analyzer API
emoji: 📊
colorFrom: blue
colorTo: gray
sdk: docker
pinned: false

📈 FinBERT: Real-Time Financial Sentiment Analysis

Machine Learning Pipeline, which analyses news headlines about finance and forecasts sentiments (Bullish, Bearish, and Neutral).

The project will train a BERT-based model with PyTorch, implement API prediction requests using FastAPI, and display visualization results on a React-Bootstrap web interface.

🔗 Project Links

• Live Application: Link
• Frontend Code: Github
• Project Blog Post: Link
• Model Weights (Optional): Hugging Face mobadara/finbert-finetuned

---

🏗️ System Architecture

• Machine Learning: PyTorch, Hugging Face transformers, Financial PhraseBank Dataset
• Backend API: Python, FastAPI, Uvicorn
• Frontend UI: React, React-Bootstrap
• Database: PostgreSQL (Neon/Supabase) via SQLAlchemy

---

🧠 The Machine Learning Pipeline

The core of this application is a fine-tuned NLP model.

1. Base Model: ProsusAI/finbert
2. Fine-tuning: Conducted in Google Colab using a T4 GPU.
3. Training Data: The Kaggle Financial PhraseBank dataset. (Check the /notebooks directory to see the complete PyTorch training loop, tokenization process, and evaluation metrics).

---

🚀 How to Run Locally

1. Clone the Repository

copy the command below and run it in your favourite terminal.

git clone https://github.com/mobadara/finbert-sentiment-analyzer-api &&
cd finbert-sentiment-analyzer-api.git

2. Create a virtual environment

python -m venv venv

3. Activate the virtual environment

a. On Linux/Mac

source venv/bin/activate

b. On Windows

venv\Scripts\activate

4. Install Dependencies

pip install -r requirements.txt

5. Start the server

uvicorn app.main:app --reload

👨‍💻 Author:

Muyiwa J. Obadara