Automatic Sarcasm Detection with Textual and Acoustic Data

Abstract

This paper takes focus on the area of automatic sarcasm detection. Automatic sarcasm detection is crucial due to the needs of sentimental analysis. The rapid development of automatic speech recognition and text mining and the large amount of voice and text data opens a broader way for researchers to open new method and improves the accuracy of automatic sarcasm detection. We observe approaches that have been used to detect sarcasm, kind of data and its features including the rises of context to improve the accuracy of automatic sarcasm detection. We found that some context cannot be reliable without the presence of other context and some approaches are very dependent on the dataset. Twitter is being used by researchers as the main mine for sentimental analysis, we notice that at some aspect it still has a flaw because it is dependent to some Twitter’s special feature that will not be found in other usual text data like hashtags and author history. Besides that, we see that the small amount of research about automatic sarcasm detection through acoustic data and its correlation with textual data could make a new opportunity in the area of sarcasm detection in speech. From acoustic data, we could get both acoustic features and textual features. Sarcasm detection with voice has the potential to get higher accuracy since it can be extracted into two data types. By describing each beneficial method, this paper could be a brief way to sarcasm detection through acoustic and textual data.