stce at SemEval-2022 Task 6: Sarcasm Detection in English Tweets

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 820 - 826 July 14-15, 2022 ©2022 Association for Computational Linguistics stce at SemEval-2022 Task 6: Sarcasm Detection in English Tweets Mengfei Yuan and Mengyuan Zhou and Lianxin Jiang and Yang Mo and Xiaofeng Shi PALI Inc. Shenzhen, China {YUANMENGFEI854, ZHOUMENGYUAN425, JIANGLIANXIN769, MOYANG853, SHIXIAOFENG309}@pingan.com.cn Abstract This paper describes the systematic approach applied in "SemEval-2022 Task 6 (iSarcas- mEval) : Intended Sarcasm Detection in En- glish and Arabic". In particular, we illustrate the proposed system in detail for SubTask-A about determining a given text as sarcastic or non-sarcastic in English. We start with the train- ing data from the officially released data and then experiment with different combinations of public datasets to improve the model general- ization. Additional experiments conducted on the task demonstrate our strategies are effective in completing the task. Different transformer- based language models, as well as some popu- lar plug-and-play proirs, are mixed into our sys- tem to enhance the model’s robustness. Further- more, statistical and lexical-based text features are mined to improve the accuracy of the sar- casm detection. Our final submission achieves an F1-score for the sarcastic class of 0.6052 on the official test set (the top 1 of the 43 teams in "SubTask-A-English" on the leaderboard). 1 Introduction Sarcasm is a sophisticated communication tech- nique to express emotions, attitudes, feelings, and evaluations. Sarcastic and ironic texts typically do not contain words with negative polarity, hostile at- titudes, or offensive in their literal sense, but rather express the contradiction or opposite of the literal meanings (Filik et al., 2016; Van Hee et al., 2018; Reyes and Rosso, 2014; Verma et al., 2021). Sar- casm detection can be considered a particular sen- timent analysis task, applied to detect texts that are intended to use some exaggeration, understatement, or rhetoric content to express criticism or praise for people or events. Many researchers have conducted different deep learning methods (Poria et al., 2016; Kumar et al., 2020; Zhang et al., 2019), traditional machine learning method (Buschmeier et al., 2014; Hernández-Farías et al., 2015; Yaghoobian et al., 2021), and big data approaches (Bharti et al., 2016; Sarsam et al., 2020; Ortega-Bueno et al., 2019) to improve the accuracy of irony or sarcasm auto- detection. SemEval-2022 Task 6 (iSarcasmEval) is a sar- casm detection task (Abu Farha et al., 2022). The standard training dataset includes 3468 English tweets and 3102 Arabic tweets. The English train- ing dataset provides 862 sarcastic tweets along with their non-sarcastic rephrases, while the Arabic datasets provides 745 sarcastic samples. For the English dataset, each sarcastic tweet is also labeled as a fine-grained multi-class and multi-label ironic tag such as satire, understatement, overstatement, and rhetorical questions. SubTask-A is a binary classification task to pre- dict whether a given tweet is sarcastic or not. Table 1 shows one sarcastic tweet and its non- sarcastic version, and one non-sarcastic tweet from the released dataset. We could notice that the raw tweets are pretty noisy and contain user …

Content truncated to 500 words