In this we worked with the Amazon Reviews Dataset. This dataset contains hundreds of thousands of reviews, each with it's raw text, it's summary and a score from 1-5 on the rating corresponding to the review. We created a multitude of discriminative models to classify the rating from the text of the review, from simple non-deep learning baseline approaches to creating multiple different deep recurrent neural networks, using attention and transfer learning to boost classification accuracy. Past that, we created a generative model to generate summaries of a given review from scratch!
Please refer the notebook itself for detailed explanation of the models and architectures tried.