Constructing and analysing the MalaySarc dataset: a resource for detecting and understanding sarcasm in Malay language

Abstract

Social media platforms provide users with an efficient and effective way to interact with content without requiring lengthy or complex textual expressions. However, sarcasm in social media discourse has become a serious problem for researchers. Compared to English and several other main languages, the research on sarcasm and the accessibility of reference materials in the Malay language are still significantly lagging. Therefore, this study aims to develop a new dataset of Malay sarcasm detection by detailing each process step, from data collection to filtering to annotation. The dataset consists of two types of data: Facebook comments and its emotion reaction buttons, which include 6,325 non-sarcastic texts and 1,380 sarcastic texts. In addition, the descriptive analysis of this dataset was also conducted to determine the usage patterns of the main features of Malay sarcasm. The analysis shows that emoji is one of the features that play an essential role in determining sarcastic comments. Besides, there are pattern-based features based on the identification of high-frequency terms in the text. The resulting dataset provides diverse examples of sarcasm that consider the linguistic and cultural nuances of the language, thus improving the accuracy and reliability of identifying social media. The findings will aid future research in developing automatic Malay sarcasm detection models using machine learning.