A Hybrid Approach for Sarcasm Detection

Abstract

There is an excessive growth in user generated textual data due to increment in internet and social media users which includes enormous amount of sarcastic words, emoji, sentences. Sarcasm is a nuanced form of communication where individual states opposite of what is implied which is done in order to insult someone, to show irritation, or to be funny. Sarcasm is considered as one of the most difficult problems in sentiment analysis due to its ambiguous nature. Recognizing sarcasm in the texts can promote many sentiment analysis and text summarization applications. So for addressing the problem of sarcasm many steps have been adopted for sarcasm detection. Different preprocessing techniques such as Hypertext markup language removal, stop words removal, etc. have been done. Similarly, conversion of the emoji and smileys into their textual equivalent has been performed. Most frequent features has been selected and a hybrid cascade and hybrid weighted average approaches which are the combinations of the algorithms random forest, naïve Bayes and support vector machine have been used for sarcasm detection. The comparison of these two approaches on different basis has been done which has shown cascade outperformed weighted approach. Moreover, comparison of cascade approaches in terms of the algorithm placement has also been performed in which random forest has proved to be the best.