This repository is an attempt to explain the predictions made by Deep Learning models. Deep Learning models are considered black box models as we often do not know why a prediction was made. With the advancements in Deep Learning it is important that we develop methods that can be used to explain the predictions made by complex models. For the purpose of explaining the predictions the shap package is used. This package is based on calculating the shapley values for the features.
The currently available techniques for explanations in NLP computes relevance of single words only. The goal of this project is to devise techniques for NLP tasks that can be used to compute the relevance of a group of words, phrases etc. The project combines the concepts of Constituency parsing with Deep neural networks to build an interpretable model. Shap is then run to generate the explanations for the predictions.
The results are quite interesting. Instead of having one word as explanations, we can see the word along with its context as explanations.
Note: This project was done as a part of XAI lab course at TUM.