Sarcasm Detection Using Features Based on Indicator and Roles

Sarcasm is a non-literalistic expression and presents a negative meaning with positive expressions. Sarcasm detection is a significant challenge for sentiment analysis which is to analyze documents with opinions. In this study, we propose a method of sarcasm detection on Twitter. We focus on two kinds of feature words. One is words modified by the indicator “ Open image in new window ”. The other is words expressing a role. First, we extract these words from tweets. Next, our method uses the lists of these words for a machine learning approach to detect sarcastic tweets. The lists of extracted words are used as features in our method. In the experiment, we compare our method with a baseline based on the features in previous studies. The experimental result shows the effectiveness of our method.