Automatic Detection of Irony

Introduction

I.1. Context and purpose

In recent years, the Internet has become a truly essential information source due to the quantity and variety of textual content available, notably, for our purposes, expressing user’s opinions. This content takes a range of forms, from blogs to comments, forums, social networks, reactions or reviews, increasingly centralized by search engines. Given the wealth of data and range of sources, there is a clear need for tools to extract, synthesize and compare extracted opinions. Tools of this type present a considerable interest for companies looking for client’s feedback on their products or brand image, and for consumers seeking guidance concerning a planned purchase, outing or trip. These tools are also of value for survey groups in evaluating market reactions to a product, for predicting the results of future elections, etc.

This is the context in which sentiment analysis, or opinion mining, emerged as a new area of research. The first work on automatic opinion extraction dates from the late 1990s, notably with Hatzivassiloglou and McKeown’s (1997) work on determining adjective polarity, and the work on document classification according to polarity (positive or negative) presented in (Pang et al. 2002) and (Littman and Turney 2002). The number of publications in this subject has increased considerably since the early 2000s, and opinion extraction is one of the most active areas in natural language processing (NLP) (Liu 2015, Benamara ...