This project contains a Jupyter Notebook which trains a model to categorise IMDB movie reviews into positive and negative sentiment.
It uses the ULMFiT approach, spelled out in this paper, to reach an accuracy of 94.4% on the Large Movie Review Dataset.