Chapter 1 Introduction

This notebook clarify how to make (supervised) machine learning models, especially NLP models interpretable. In order to understand this notebook, the readers are expected to prequire some basic knowledge of machine learning and NLP. If you are new to machine learning and NLP, I recommend you to watch Andrew Ng’s “Machine Learning” online course and Andrew Ng’s “Natural Language Processing” online course. This notebook contains some mathematical formulas, but the goal is to explain the concepts rather than to prove theorems.

It would be too ambitious to cover all the topics related to interpretability of machine learning models in a single notebook since the techniques are diverse and the field is still evolving. However, it is essential to understand the basic concepts of interpretability and methods so that similar ideas could be applied to other models. The scope of this notebook is interpretable NLP models dedicated to text processing.

The first chapter talks about the overview of the topics in explainable AI. It starts with an introduction of interpretability and explainability in machine learning with daily life examples. I will discuss why interpretability is important by exploring what type of users are looking for interpretability for diverse goal, and what types of questions could be answered. I will discuss different taxonomies of interpretable machine learning methods use one to group the chapters about XNLP methods. Next section is to evaluate the interpretability of machine learning including what and how they are to be evaluated. The last section of this chapter is about plugging in social science insights to explain what a good explanation is.

To concrete the method concepts, I will use the real-world dataset presented in the Data chapter.

The interpretable NLP methods are divided into several chapters including Intrinsic Interpretable Models, Global Methods and Local Methods.

The last chapter is about the challenge and the future of interpretable NLP.

Glossary used throughout the book can be looked up in the Glossary. You can either read the book from beginning to end or jump directly to the methods that interest you. I hope you will enjoy the read!