Automatic Sentiment Analysis

This repository contains code for an automatic sentiment analysis project using image and text data. The project uses a Variational Autoencoder (VAE) to combine image and text embeddings for sentiment analysis.

Project Structure

├── best_vae_model.pth
├── environment.yml
├── inference_output/
├── README.md
├── scripts/
│   └── flickr_test/
│       ├── 0_flickr_dataset_setup.py
│       ├── 1_flickr_BERT_featurisation.py
│       ├── 2_create_training_validation_PCA.py
│       ├── 2_create_training_validation_tSNE.py
│       ├── 3_test_train.py
│       ├── 4_inference.py
│       └── models_and_loaders.py
├── sentiment_AE/
│   ├── models.py
│   ├── settings.py
│   └── training.py
├── setup.sh
├── train_val_split_visualization_pca.png
└── train_val_split_visualization_tsne.png

Installation

1. Clone this repository:

   git clone https://github.com/yourusername/automatic_sentiment.git
   cd automatic_sentiment
   

2. Set up the Conda environment:

   chmod +x setup.sh
   ./setup.sh
   

   Alternatively, you can create the environment from the environment.yml file:

   conda env create -f environment.yml
   

3. Activate the Conda environment:

   conda activate automatic_sentiment
   

Data Preparation and Processing

1. Set up the Flickr dataset:

   python scripts/flickr_test/0_flickr_dataset_setup.py
   

2. Generate BERT features for the dataset:

   python scripts/flickr_test/1_flickr_BERT_featurisation.py
   

3. Create training and validation splits:

   python scripts/flickr_test/2_create_training_validation_PCA.py
   

   or

   python scripts/flickr_test/2_create_training_validation_tSNE.py
   

Training

To train the VAE model:

python scripts/flickr_test/3_test_train.py

You can modify hyperparameters in the sentiment_AE/settings.py file.

Inference

Note: Working model weights are given by default - please get in touch if this is giving strange outputs.

To run inference on new data, use the following command:

python scripts/flickr_test/4_inference.py --image_dir raw_data/skyscrapers --text_list 'love' 'hate'

This command will:

• Use images from the raw_data/skyscrapers directory
• Apply the sentiments 'love' and 'hate' to each image
• Output the results in the inference_output directory

You can modify the --image_dir and --text_list arguments to use different input images and sentiments.

Project Components

• scripts/flickr_test/models_and_loaders.py: Contains model definitions and data loaders
• sentiment_AE/models.py: Defines the VAE model architecture
• sentiment_AE/settings.py: Contains project settings and hyperparameters
• sentiment_AE/training.py: Implements the training loop
• best_vae_model.pth: The best trained VAE model
• train_val_split_visualization_pca.png and train_val_split_visualization_tsne.png: Visualizations of the train-validation split using PCA and t-SNE

Contributing

Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests.

License

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

Acknowledgments

• Flickr dataset providers
• Contributors to PyTorch and Transformers libraries