Binarized Word Mover's Distance

For my master’s thesis I developed an alternative lower bound to an existing textual distance metric which leverages binary encoded word embedding vectors to reduce memory requirements while maintaining downstream accuracy relative to pre-existing lower bounds. This new lower bound is demonstrably performant but warrants further evaluation. I presented this paper at Repl2NLP-2022. During this project I gained more skill working with TensorFlow, as this framework was used to build the autoencoder for producing binary word vectors. I also improved my Python coding skills with respect to object-oriented programming and developed the ability to evaluate a methodology with academic rigour. For further information please refer to the PyPI page or GitHub repository.