IMDB Sentiment Analysis with a simple neural network

Project Overview

This project implements a sentiment analysis model using a simple neural network trained on the IMDB movie review dataset. It demonstrates the process of loading and preprocessing text data, building a basic neural network model, and training it for binary sentiment classification (positive/negative reviews).

Model Architecture

The sentiment analysis model uses a simple feed-forward neural network with the following structure:

1. Input Layer: Accepts tokenized text input
2. Embedding Layer: Converts tokens to dense vector representations
3. Hidden Layer 1: Fully connected layer with ReLU activation
4. Dropout: Applies dropout for regularization
5. Hidden Layer 2: Fully connected layer with ReLU activation
6. Output Layer: Produces the final sentiment prediction

Features

• Data loading and preprocessing using the Hugging Face datasets library
• Simple neural network model implemented in PyTorch
• Training loop with early stopping and learning rate scheduling

Setup

1. Clone the repository:

   git clone https://github.com/yourusername/sentiment-analysis.git
   

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`
   

3. Install the required dependencies:

   pip install -r requirements.txt
   

Usage

To train the model with default parameters, simply run:

python train_with_simple_nn.py

You can customize the training process by passing command-line arguments:

python train_with_simple_nn.py --max_length 256 --train_size 25000 --batch_size 64 --embedding_dim 200 --hidden_dim 128 --dropout_rate 0.3 --learning_rate 0.0005 --patience 7

Model Architecture

The sentiment analysis model is a simple feed-forward neural network with the following structure:

1. Embedding layer
2. Two hidden layers with ReLU activation
3. Output layer

Dropout is applied after each hidden layer for regularization.

Results

After training, the model's performance will be displayed, showing the test accuracy. A plot of the training history (loss and accuracy for both training and validation sets) will be saved as training_history.png.

Future Improvements

• Experiment with more complex architectures (e.g., LSTM, Transformer)
• Implement cross-validation for more robust evaluation
• Add support for loading pre-trained word embeddings
• Extend the model to handle multi-class classification
• Implement a simple web interface for real-time sentiment analysis

Contributing

Contributions to this project are welcome! Please fork the repository and submit a pull request with your proposed changes.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• The IMDB dataset provided by Stanford AI Lab
• Hugging Face for their datasets and transformers libraries
• PyTorch team for their excellent deep learning framework

Contact

For any questions or feedback, please open an issue in the GitHub repository.