Indonesian Sarcasm Detection Using Convolutional Neural Network

Abstract

In sentiment analysis, sarcasm is considered to be one of the most difficult problems to solve. Sarcasm or verbal irony is a linguistic phenomenon where the wording of a speaker signifies the opposite of the literal meaning of what they are saying. On social media, Indonesians tend to use sarcastic remarks to criticize something and this behavior can pose a problem when a sentiment analysis is performed where the analysis yields false-positive results. To solve this potential problem, the detection of sarcasm is necessary. A convolutional neural network model is used in this paper to handle the automated identification of sarcasm in Indonesian. Although it is mainly used in computer vision, this model has shown promising results on specific kinds of sentiment analysis and classification problems compared to other neural network models. Using this model can increase detection accuracy compared to discrete models. There have been many studies conducted on sarcasm detection, although the majority is in English. Other languages such as Indonesian are slow in improvement compared to English. This fact can be seen by the lack of publications on Indonesian sarcasm detection studies. This study is conducted to contribute to other Indonesian sarcasm detection studies.