Integrated Gradients for Natural Language Processing from scratch
Open the deep neural network black box, and visualize feature importance regardless of model architecture. Implementation of ideas from the paper Axiomatic Attribution for Deep Networks on text data
