Sarcasm Detection in Indonesian Tweets Using Hyperbole Features

Abstract

Since sarcasm has inverse meaning from what is said or written, it is very hard to detect sarcasm. Therefore, detecting sarcasm is an important task in Natural Language Processing (NLP) field. In this study, we use interjection, intensifier, capital letters, elongated words, and punctuation marks as hyperbole features to detect sarcasm in Indonesian tweets. Particularly, these hyperbole features are utilized by Support Vector Machine (SVM), Random Forest (RF), and RF+Bagging to classify Indonesian tweets in our testing data as sarcasm or not-sarcasm. English tweets obtained from Kaggle and SemEval are employed as our training data, while Indonesian tweets obtained from Drone Emprit are used as the testing data. Our experimental results show that our model with hyperbole features classifies more the tweets in the testing data as sarcasm than that without hyperbole ones. Our observation indicates that using hyperbole features could contribute well to detecting sarcasm.