LISACTeam at SemEval-2022 Task 6: A Transformer based Approach for Intended Sarcasm Detection in English Tweets

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 993 - 998 July 14-15, 2022 ©2022 Association for Computational Linguistics LISACTeam at SemEval-2022 Task 6: A Transformer based Approach for Intended Sarcasm Detection in English Tweets Abdessamad Benlahbib1∗, Hamza Alami2∗, Ahmed Alami3 1 LISAC Laboratory, Faculty of Sciences Dhar EL Mehraz, USMBA, Fez, Morocco 2 Innov-Tech Laboratory, Department of Engineering, High Technology School, Rabat 3 Laboratory of Engineering Sciences, National School of Applied Sciences, Ibn Tofail University, Kenitra, Morocco abdessamad.benlahbib@usmba.ac.ma , hamza0alami@gmail.com alami.alami1996@gmail.com Abstract In this paper, we present our system and find- ings for SemEval-2022 Task 6 - iSarcasmEval: Intended Sarcasm Detection in English. The main objective of this task was to identify sarcastic tweets. This task was challenging mainly due to (1) the small training dataset that contains only 3468 tweets and (2) the imbal- anced class distribution (25% sarcastic and 75% non-sarcastic). Our submitted model (ranked eighth on Sub-Task A and fifth on Sub-Task C) consists of a Transformer-based approach (BERTweet model). 1 Introduction The Cambridge Dictionary 1 defines sarcasm as "the use of remarks that clearly mean the opposite of what they say, made in order to hurt someone’s feelings or to criticize something in a humorous way." Due to the Web openness, sarcastic content becomes very frequent in social media and e- commerce platforms, which may cause misunder- standings. Furthermore, identifying such content is a very challenging task even for humans (Farias and Rosso, 2017). Also, it could impact some nat- ural language processing tasks such as sentiment analysis (Farias and Rosso, 2017; Do et al., 2019; Tubishat et al., 2018; Balazs and Velásquez, 2016; Maynard and Greenwood, 2014; Ptáˇcek et al., 2014; Bouazizi and Otsuki Ohtsuki, 2016; Ren et al., 2018). We introduce the following example: "The movie was enjoyable to the point that I clapped because it is finished." For an opinion mining system, this sentence could be considered positive. However, the author expresses a negative judgment against the movie since the expression "I clapped because it is finished" means that it was boring. ∗contributed equally 1https://dictionary.cambridge.org/fr/ dictionnaire/anglais/sarcasm For this reason, SemEval 2022 set up Task 6: iSarcasmEval - Intended Sarcasm Detection in English and Arabic to detect sarcastic and non- sarcastic tweets (Abu Farha et al., 2022). Our submitted system consisted of a pre-trained trans- former model for English Tweets named BERTweet (Nguyen et al., 2020), secured 8th and 5th positions respectively on Sub-Task A and Sub-Task C leader- board. The rest of the paper is structured in the follow- ing manner: Section 2 provides the data structure and the main objective of each Sub-Task. Section 3 describes our system. Section 4 details the ex- periments. And finally, Section 5 concludes this paper. 2 Task Description The organizers of this task introduced two tweet datasets for both English and Arabic languages that contain: • a label specifying whether a tweet is sarcastic or non-sarcastic, provided by its author. • a non-sarcastic rephrase of a sarcastic tweet provided by its author. …

Content truncated to 500 words