I2C at SemEval-2022 Task 6: Intended Sarcasm Detection on Social Networks with Deep Learning

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 877 - 880 July 14-15, 2022 ©2022 Association for Computational Linguistics Abstract In this paper we present our approach and system description on iSarcasmEval: a SemEval task for intended sarcasm detection on social networks. This derives from our participation in SubTask A: Given a text, determine whether it is sarcastic or non-sarcastic. In our approach to complete the task, a comparison of several machine learning and deep learning algorithms using two datasets was conducted. The model which obtained the highest values of F1-score was a BERT-base-cased model. With this one, an F1-score of 0.2451 for the sarcastic class in the evaluation process was achieved. Finally, our team reached the 30th position. 1 Introduction Sarcasm is a form of mockery intended to imply the opposite or to express displeasure. It can be classified as irony, satire, understatement, overstatement, rhetorical question and sarcasm itself. Social media platforms allow people to express themselves and speak about several different topics via text, emojis and multimedia. Many companies collect information about people’s opinions to aid in marketing decision- making about their products. Sarcasm detection is relevant in data analysis because it can be wrongly interpreted due to its nature, becoming potentially harmful to different computational systems. For example, if a computer system understands the sentence “The only thing I got from college is a caffeine addiction”, its literal meaning would not be understood, as it refers to how stressful college can be. Defining a phrase as sarcastic is a really challenging task to solve for the text mining community because there are many details that need to be considered such as the writer’s personality and the sentence context. For instance, depending on the ideology of the person, “Trump, that respectful man” could be sarcastic or not. In iSarcasmEval: Task 6, SubTask A (Abu Farha et al. 2022) participants had to decide if a particular tweet includes a sarcastic connotation or not. 2 Background Two sample datasets were used to train the model. The first one was the original dataset without any preprocessing. This dataset is composed of 3467 rows, with 2600 0-value rows (non- sarcastic) and 867 1-value rows (sarcastic). The second sample dataset (Tweet – Rephrase) consists of a phrase and rephrase contrast in which the aim was to adequately learn the differences between a phrase with a sarcastic connotation and that same sentence expressing the same thing but in a literal way. This dataset is composed of 1734 rows: 867 1-value rows and 867 0-value rows, where these last ones are rephrases related to each tweet taken as a 1-value row. So, these two datasets follow the same structure: - tweet: “See Brexit is going well” - sarcastic: “1” I2C at SemEval-2022 Task 6: Intended Sarcasm Detection on Social Networks with Deep Learning Pablo González Díaz Escuela Técnica Superior de Ingeniería Universidad de Huelva (Spain) pablo.gonzalez682@alu.uhu.es Pablo Cordón Hidalgo Escuela Técnica Superior de Ingeniería Universidad de Huelva (Spain) pablo.gonzalez682@alu.uhu.es Jacinto Mata Vázquez …

Content truncated to 500 words