Amrita_CEN at SemEval-2022 Task 6: A Machine Learning Approach for Detecting Intended Sarcasm using Oversampling

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 834 - 839 July 14-15, 2022 ©2022 Association for Computational Linguistics Amrita_CEN at SemEval-2022 Task 6: A Machine Learning Approach for Detecting Intended Sarcasm using Oversampling Aparna K Ajayan, Krishna Mohanan, Anugraha S, Premjith B, and Soman K.P Centre for Computation Engineering and Networking (CEN) Amrita School of Engineering, Amrita Vishwa Vidyapeetham, India b_premjith@cb.amrita.edu Abstract This paper describes the submission of the team Amrita_CEN to the shared task on iSarcasm Eval: Intended Sarcasm Detection in English and Arabic at SemEval 2022. The sarcasm de- tection task was formulated as a classification problem and modelled using machine learn- ing classifiers. We used K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Naïve Bayes, Logistic Regression, Decision Tree, and the Random Forest ensemble method. In addi- tion, the class imbalance problem in the dataset was addressed using a feature engineering tech- nique. We submitted the predictions by SVM, Logistic Regression and Random Forest ensem- ble based on the performance during training. 1 Introduction Sarcasm is an ironic form showing a disparity be- tween the actual and intended meaning of the text affecting the decision-making process. These are reflected in our day-to-day communication with each other happening in social media forums. Twit- ter exhibits rich sarcasm phenomena, thereby en- couraging automatic sarcasm detection methods and removing such tweet data. Due to the sociocul- tural aspects of sarcastic communication, the major- ity of the sarcasm detection work has been focused only on the English language (Oprea and Magdy, 2020b), and only a limited amount of work was done in other languages such as Arabic (El Mah- daouy et al., 2021). Identification of sarcastic comments from social media contexts is essential since the author and the receiver are at various places. Therefore, ex- changing conversations may sometimes lead to a negative meaning of the text that even the author has not meant to convey. Moreover, the data stream for sarcasm does not exhibit any static structure like specific tags in the form of #sarcasm, and #irony (Ptáˇcek et al., 2014) (Khodak et al., 2018). This event can lead to noisy labels due to several rea- sons, as outlined by (Oprea and Magdy, 2020b). Other works reported on the topic mainly depend on manual labelling, provided with manually anno- tated sarcasm labels. In (Oprea and Magdy, 2020b) the authors pointed out that manual labelling rep- resents the author annotation in contrast with the intention of the authors. The sarcasm prediction on Twitter that influences Machine Intelligence is a challenging task (Khare et al., 2022). It can be achieved with the help of the Natural Language Processing (NLP) approach, and many recent works on automatic sarcasm de- tection have focused on Twitter data as it primarily requires an understanding of the human expres- sions, language, and emotions expressed via tex- tual or non-textual content (Kumar et al., 2021). Therefore, the goal of the SemEval shared task is to facilitate the development of machine learning models that can detect …

Content truncated to 500 words