Beyond Face Value: Identifying Irony and Sarcasm in Short Texts Through Multiclass Classification Techniques

This research study presents a comparative analysis of multiclass classifiers for detecting irony and sarcasm in short texts. With the increasing use of social media, it has become important to develop accurate and efficient algorithms for detecting irony and sarcasm, which are often expressed implicitly in text. This study has evaluated five multiclass classifiers, including Naive Bayes, linear classifier, XG Boost, KNN, and SGD, using different text representations, such as count vectors, word-level TF-IDF, and hash vectors. The results showed that the linear classifier using word-level TF-IDF achieved the highest accuracy of 74.39%, while the KNN classifier using hash vectors achieved the highest accuracy of 75.16%. However, all classifiers exhibited sensitivity to certain keywords and phrases, indicating the need for further research to improve the robustness of the classifiers. This study provides insights into the strengths and weaknesses of different multiclass classification approaches for detecting irony and sarcasm in short texts, which can guide future research in this field.