Reimplementation and experimentation with the ideas presented in Rationalizing Neural Predictions. This project seeks to perform multi-sentiment analysis on free-form text data. In addition, the model provides an extract from the original text on why it predicted the sentiment it did. In essence, it predicts sentiement accurately and provides a rationale for why it predicted the way it did.
Report describing our project and findings found here
Main file containing our model architecture in generator.py
supporting data prepocessing and postprocessing in preprocess.py and postprocess.py.