Class-Activation-Maps-in-NLP

This repo is part of the work I did in my summer internship at CNeRG Lab, IIT Kharagpur.

Description

The idea is based on usage of class activation maps to understand the words influencing the prediction of a CNN model trained on the SNLI dataset. The output classes consist of Entailment, Neutral and Contradiction. Class Activation maps have been used for images prior with the following framework: I have used a 100 dimensional Glove embedding and premise and hypothesis padded to 30 words accordingly and concatenated.