Skip to content

README.md

Latest commit

 

History

History
127 lines (101 loc) · 3.76 KB

README.md

File metadata and controls

127 lines (101 loc) · 3.76 KB
Raw

Earnings Call Sentiment Analyzer

A production-grade NLP system that scores sentiment from earnings call transcripts using a fine-tuned Longformer model and backtests whether sentiment predicts post-earnings price moves.

Python FastAPI React HuggingFace License

Overview

This project fine-tunes allenai/longformer-base-4096 on labeled earnings call transcripts, exposes a FastAPI inference server, and provides a React dashboard for real-time analysis and event-study backtesting.

Input: Earnings call transcript text
  └─> Longformer tokenizer (max 4096 tokens)
      └─> Fine-tuned classifier head
          └─> Sentiment class + confidence + key phrases
              └─> Event-study backtest vs yfinance price data

Features

  • Fine-tune Longformer on custom earnings transcript dataset
  • REST API with FastAPI for real-time inference
  • Sentiment scoring: Positive / Neutral / Negative
  • Key phrase extraction with attention visualization
  • Event-study backtester using yfinance price data
  • React + TypeScript dashboard with charts
  • Docker Compose for one-command deployment
  • Full test suite with pytest

Project Structure

earnings-sentiment-analyzer/
├── backend/
│   ├── app/
│   │   ├── api/          # FastAPI route handlers
│   │   ├── core/         # Config, logging, security
│   │   ├── models/       # Pydantic schemas
│   │   ├── services/     # NLP, backtest, price data
│   │   └── utils/        # Helpers
│   └── tests/
├── frontend/
│   └── src/
│       ├── components/
│       ├── hooks/
│       ├── services/
│       └── store/
├── notebooks/            # Training & EDA notebooks
├── data/                 # Raw and processed datasets
├── scripts/              # Training, eval, data prep
└── docs/

Quickstart

With Docker

git clone https://github.com/yourusername/earnings-sentiment-analyzer
cd earnings-sentiment-analyzer
cp .env.example .env
docker-compose up --build

App: http://localhost:5173
API docs: http://localhost:8000/docs

Manual Setup

# Backend
cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

# Frontend
cd frontend
npm install && npm run dev

Train the Model

cd scripts
python train.py --model allenai/longformer-base-4096 \
                --data ../data/processed/transcripts.csv \
                --epochs 5 --batch-size 4 --lr 2e-5

API Reference

Method Endpoint Description
POST /api/v1/analyze Analyze a transcript
POST /api/v1/backtest Run event-study backtest
GET /api/v1/health Health check
GET /api/v1/model/info Model metadata

Model Performance

Class Precision Recall F1 Support
Negative 0.43 0.46 0.45 218
Neutral 1.00 0.00 0.00 89
Positive 0.46 0.59 0.51 293
Weighted Avg 0.53 0.45 0.41 600

Tech Stack

Layer Technology
ML Model HuggingFace Transformers, PyTorch
Backend FastAPI, Pydantic, SQLAlchemy
Frontend React 18, TypeScript, Recharts, Zustand
Price Data yfinance
Database PostgreSQL (prod), SQLite (dev)
Containerization Docker, Docker Compose
Testing pytest, pytest-asyncio

License

MIT