Comparison of Feature Extraction for Sarcasm on Twitter in Bahasa

This study aims to detect the text of sarcasm in Bahasa. Sarcasm detection is very important in the field of affective computing and sentiment analysis because expressions of sarcasm can reverse the polarity of sentences. Sarcasm is difficult to detect in text because there is no intonation of sounds and facial expressions. Therefore, in this study, a system is created to recognize the sentence of sarcasm in text. The data consist of 480 train data and 120 test data collected by crawling on Twitter. Then, the data passed through the preprocessing and feature extraction stages. Classification of sarcasm and non-sarcasm sentences uses the Support Vector Machine (SVM) algorithm. Experiments are done by comparing the accuracy of N-gram, POS Tag, Punctuation, Pragmatic and combining all features. Our proposed approach reaches the highest accuracy of 91.6% with a precision of 92% when all features are combined.