Introduction

Let's say you're working with product reviews for a mobile phone and your task is to classify the sentiment in the reviews as being positive or negative. You encounter a review that says: "The phone does not have a great camera, or an amazingly vivid display, or an excellent battery life, or great connectivity, or other great features that make it the best." Now, when you read this, you can easily identify that the sentiment in the review is negative, despite the presence of many positive phrases such as "excellent battery life" and "makes it the best". You understand that the presence of the term "not" right toward the beginning of the text negates everything else that comes after.

Will the models we've created so far be able to identify the sentiment in such a case? Probably not, because if your models don't realize that the term "not" toward the beginning of the sentences is important and needs...