Assessment of Supervised Learning Algorithms for Irony Detection in Online Social Media

Abstract

The social media has become one of the most widely used communication tools for people to share their opinions with others. The irony, with a simple definition, is the creative use of language, and it has recently attracted the attention of computer scientists. There is a vast amount of raw data, collected from social media resources, containing ironical statements. It is very difficult to analyze the rapidly growing data manually. Additionally, the character size limitations and typographical errors of online social media tools make the traditional methods of classification insufficient for detection task. In this study, the detection of irony in online social media is modeled as a classification problem, and the success of supervised machine learning methods in real data is assessed. Bayesian Network (BayesNet), OneR, Stochastic Gradient Descent (SGD), Logistic Model Tree (LMT), Multi-Layer Perceptron (MLP), Radial Basis Function Networks (RBF), Voted Perceptron, IBk, Randomizable Filtered Classifier (RFC), Isolation Forest, Fuzzy Lattice Reasoning (FLR), and Bagging algorithms are applied for the first time on irony detection problem and a comprehensive evaluation is provided. © 2019 IEEE.