Irony Detection with Deep Learning in Turkish Microblogs

Abstract

The number of people sharing on social media is constantly increasing. One of the most popular microblogging sites is Twitter, where 500 million tweets are posted every day. Categorizing manually in such large data is a challenging task. Therefore, classification using autonomous systems is of great importance. Irony is a term in which the opposite meaning of something is meant. The verbal or spoken verb, under its serious appearance, aims to speak opposite or to draw the verb to the point of contradiction. Recently, after the successful results of emotion analysis over the tweets, studies on the determination of irony have been made. While it is easier to detect irony in face-to-face conversation, it can be difficult even for normal people to understand it in written communication. The character limit on Twitter prevents some people from applying classification methods to spelling mistakes and carelessness in punctuation. For this reason, it has become obligatory to perform preprocessing steps in the first step. After correcting the data, machine learning and deep learning algorithms were applied with different parameters and the success of the results were examined and compared.