AlexU-AL at SemEval-2022 Task 6: Detecting Sarcasm in Arabic Text Using Deep Learning Techniques

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 891 - 895 July 14-15, 2022 ©2022 Association for Computational Linguistics AlexU-AL at SemEval-2022 Task 6: Detecting Sarcasm in Arabic Text Using Deep Learning Techniques Aya Lotfy, Marwan Torki and Nagwa El-Makky Computer and Systems Engineering Department Alexandria University Alexandria, Egypt eng-aya.lotfy1419, mtorki, nagwa.elmakky@alexu.edu.eg Abstract Sarcasm detection is an important task in Nat- ural Language Understanding. Sarcasm is a form of verbal irony that occurs when there is a discrepancy between the literal and intended meanings of an expression. In this paper, we use the tweets of the Arabic dataset provided by SemEval-2022 task 6 to train deep learn- ing classifiers to solve the sub-tasks A and C associated with the dataset. Sub-task A is to determine if the tweet is sarcastic or not. For sub-task C, given a sarcastic text and its non-sarcastic rephrase, i.e. two texts that con- vey the same meaning, determine which is the sarcastic one. In our solution, we utilize fine- tuned MARBERT (Abdul-Mageed et al., 2021) model with an added single linear layer on top for classification. The proposed solution achieved 0.5076 F1-sarcastic in Arabic sub-task A, accuracy of 0.7450 and F-score of 0.7442 in Arabic sub-task C. We achieved the 2nd and the 9th places for Arabic sub-tasks A and C respectively. 1 Introduction Sarcasm is ubiquitous phenomenon on the social web, and is difficult to be analysed automatically and manually by humans because of its nature. Sarcasm data can be very confusing to computer systems which use it to perform tasks such as sentiment analysis, opinion mining, author profil- ing, and harassment detection (Liu, 2012; Rosen- thal et al., 2014; Maynard and Greenwood, 2014; Van Hee et al., 2018). (Rosenthal et al., 2014) show that the senti- ment polarity classification performance on non- sarcastic tweets is much better than on sarcastic ones, in the context of SemEval. Sentiment polar- ity classification is used widely in industry, driv- ing marketing, administration, and investment de- cisions (Hassan Yousef et al., 2014). So it is impor- tant to create models for sarcasm detection. A comparatively small dataset is a challenge we faced when working on the Arabic dataset that makes it difficult to train complex models. We used transfer learning to treat this issue. By using a transfer learning, a pre-trained model for some task on a large dataset can be used as a starting point in another task which improves the performance. It is used in a wide range of natural language processing (NLP) tasks. Word embeddings such as word2vec (Mikolov et al., 2013), FastText (Joulin et al., 2016) and Glove (Pennington et al., 2014) can be used to ini- tialize vectors learnt form large dataset. Recently, Bidirectional Encoder Representations from Trans- formers (BERT) (Devlin et al., 2018) became the most popular NLP approach to transfer learning. Google AI Language team pretrained BERT model and fine-tuned it for a large range of tasks, such as question answering and language inference where it achieved …

Content truncated to 500 words