Unsupervised Style Transfer: Automatic Sentiment Transfer Using Classification Attention Weights

Code for my Master's Thesis Information Science at the University of Groningen.

How to run the models

First preprocess the data using: preprocess.py

Before we can train the HAN model we first need to get the POS-tags using: HAN/POS.ipynb

Next we can train the HAN model using: HAN/HAN+POS_attention_mechanism.ipynb

We can then use the style_generator.ipynb to generate sentences from one sentiment to the opposing sentiment

We can use train_evaluation.ipynb to train the classifiers for automatic classification

Lastly, we can use evaluation.ipynb and all_evaluation.ipynb to evaluate the human and automatic evaluations

Note:
- This is research code and might therefore not be fully complete. 
- For questions and full results contact the author.

Authors

• Chi Sam Mac - Github - E-Mail - LinkedIn

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details

Acknowledgments

• M. Nissim for mentoring my project giving me guidance and tips