IMDb RNN Genre Classifier — Backend + Frontend

This project wraps the trained model from IMDB_RNN_Project.ipynb in a FastAPI backend and Streamlit frontend.

Model used

The notebook uses:

• Tokenizer(num_words=5000)
• pad_sequences(..., maxlen=200)
• Embedding(input_dim=5000, output_dim=64)
• SimpleRNN(64)
• Dense(32, relu)
• Dense(202, softmax)

The notebook trains for 5 epochs with batch size 32 and reports test accuracy of about 7.5%. The target is movie genre, not positive/negative sentiment.

Project structure

imdb_rnn_web_app/
├── backend/
│   ├── main.py
│   ├── prepare_artifacts.py
│   ├── requirements.txt
│   ├── tokenizer.pkl          # generated
│   └── label_encoder.pkl      # generated
├── frontend/
│   ├── app.py
│   └── requirements.txt
├── model/
│   └── imdb_rnn_model.keras
├── imdb_dataset.csv           # place your original dataset here
└── README.md

Important

The .keras model does not contain the Keras Tokenizer or sklearn LabelEncoder from the notebook. Therefore, generate those artifacts from the same imdb_dataset.csv used in training.

The preprocessing in prepare_artifacts.py matches the notebook:

• lowercase text
• remove non A-Z characters
• Tokenizer(num_words=5000)
• fit tokenizer on Overview
• fit LabelEncoder on Genre

Run backend

From the project root:

pip install -r backend/requirements.txt
python backend/prepare_artifacts.py
uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000

Then open:

http://127.0.0.1:8000/docs

Run frontend

Open a second terminal:

pip install -r frontend/requirements.txt
streamlit run frontend/app.py

The Streamlit app sends the overview to:

POST http://127.0.0.1:8000/predict

API example

Request:

{
  "overview": "A detective investigates a mysterious murder in a small town."
}

Response:

{
  "genre": "Crime, Drama",
  "confidence": 0.12,
  "top_predictions": [
    {
      "genre": "Crime, Drama",
      "confidence": 0.12
    }
  ]
}

The exact predicted values depend on the trained model and input text.