Abstract
Recognizing both literal and figurative meanings is crucial to understanding users' opinions on various topics or events in social media. Detecting the sarcastic posts on social media has received much attention recently, particularly because sarcastic comments in the form of tweets often include positive words that represent negative or undesirable characteristics. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement was used to understand the application of different machine learning algorithms for sarcasm detection in Twitter. Extensive database searching led to the inclusion of 31 studies classified into two groups: Adapted Machine Learning Algorithms (AMLA) and Customized Machine Learning Algorithms (CMLA). The review results revealed that Support Vector Machine (SVM) was the best and the most commonly used AMLA for sarcasm detection in Twitter. In addition, combining Convolutional Neural Network (CNN) and SVM was found to offer a high prediction accuracy. Moreover, our result showed that using lexical, pragmatic, frequency, and part-of-speech tagging can contribute to the performance of SVM, whereas both lexical and personal features can enhance the performance of CNN-SVM. This work also addressed the main challenges faced by prior scholars when predicting sarcastic tweets. Such knowledge can be useful for future researchers or machine learning developers to consider the major issues of classifying sarcastic posts in social media.