Auto Author Recognition

A web application for automatic author recognition from text using deep learning models (LSTM and Transformer). The app provides a FastAPI backend for inference and a Streamlit frontend for user interaction.

---

Features

• LSTM and Transformer Models: Predicts the author of a given text using either an LSTM-based or Transformer-based classifier.
• REST API: FastAPI endpoints for model inference.
• Web UI: Streamlit frontend for uploading text, selecting models, and viewing predictions.
• Docker Support: Easily run the entire stack in a container.

---

Architecture

• FastAPI Backend (app/main.py): Serves /predict/lstm/ and /predict/transformer/ endpoints.
• LSTM Module (app/lstm/lstm_inference.py): Loads LSTM model and weights for inference.
• Transformer Module (app/transformer/transformer_inference.py): Loads Transformer model and weights for inference.
• Streamlit Frontend (app/streamlit_app.py): User interface for uploading text and viewing predictions.

---

Getting Started

Prerequisites

• Python 3.10+
• pip
• (Optional) Docker

Installation (Local)

1. Clone the repository:

   git clone <repo-url>
   cd auto-author-recognition
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Download/Place Model Weights:

   • LSTM weights: app/lstm/weights/checkpoint_epoch3794_valacc0.72.pt
   • Transformer weights: Download the weights folder from the provided Google Drive link and place it inside app/transformer/ so that you have app/transformer/weights/ containing the model files.

4. Run FastAPI backend:

   uvicorn app.main:app --reload --port 8000
   

5. Run Streamlit frontend (in a new terminal):

   streamlit run app/streamlit_app.py
   

Using Docker

1. Build the Docker image:

   docker build -t auto-author-recognition .
   

2. Run the container:

   docker run -p 8000:8000 -p 8501:8501 auto-author-recognition
   

---

Usage

• Web UI:

  • Open http://localhost:8501 in your browser.
  • Upload or paste text, select a model, and click "Predict Author".

• API Endpoints:

  • POST /predict/lstm/ with JSON { "text": "..." }
  • POST /predict/transformer/ with JSON { "text": "..." }

  Example using curl:

  curl -X POST "http://localhost:8000/predict/lstm/" -H "Content-Type: application/json" -d '{"text": "your text here"}'
  

---

File Structure

app/
  main.py                # FastAPI app
  streamlit_app.py       # Streamlit UI
  lstm/
    lstm_inference.py    # LSTM model logic
    weights/             # LSTM model weights
  transformer/
    transformer_inference.py # Transformer model logic
    weights/                 # Transformer model files
requirements.txt
Dockerfile
architecture.dot         # Architecture diagram (Graphviz)

---

Notes

• Make sure the model weights are present in the correct folders before running.
• The app uses the aubmindlab/araelectra-base-discriminator model for embeddings.
• For production, update CORS and security settings as needed.

---

License

MIT License