JUST_ONE at SemEval-2023 Task 10: Explainable Detection of Online Sexism (EDOS)

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 526–531 July 13-14, 2023 ©2023 Association for Computational Linguistics JUST_ONE at SemEval-2023 Task 10: Explainable Detection of Online Sexism (EDOS) Doaa Obeidat, Wala’a Shnaigat, Heba Nammas, Malak Abdullah Jordan University of Science and Technology, Irbid, Jordan dfobeidat20, washnaigat20, hrnammas20@cit.just.edu.jo, mabdullah@just.edu.jo Abstract The problem of online sexism, which refers to offensive content targeting women based on their gender or the intersection of their gen- der with one or more additional identity char- acteristics, such as race or religion, has be- come a widespread phenomenon on social me- dia. This can include sexist comments and memes. To address this issue, the SemEval- 2023 international workshop introduced the "Explainable Detection of Online Sexism Chal- lenge", which aims to explain the classifica- tions given by AI models for detecting sexism. In this paper, we present the contributions of our team, JUST_ONE, to all three sub-tasks of the challenge: subtask A, a binary classifica- tion task; subtask B, a four-class classification task; and subtask C, a fine-grained classifica- tion task. To accomplish this, we utilized pre- trained language models, specifically BERT and RoBERTa from Hugging Face, and a se- lective ensemble method in task 10 of the Se- mEval 2023 competition. As a result, our team achieved the following rankings and scores in different tasks: 19th out of 84 with a Macro-F1 score of 0.8538 in task A, 22nd out of 69 with a Macro-F1 score of 0.6417 in task B, and 14th out of 63 with a Macro-F1 score of 0.4774 in task C. Keywords: Sexism, Muslim woman, Fine- grain, NLP, Transformer, Hugging face, Text classification 1 Introduction Social media has facilitated the freedom of ex- pression, but it has also enabled the spread of hate speech against specific communities, such as races, sexes, and ethnicities (Zhang and Luo, 2019; Habash et al., 2022; Faraj and Abdullah, 2021). Sexism is a particularly challenging form of hate speech to address because it can be difficult to identify and significantly impacts women’s lives. Women are often subjected to abusive and discrimi- natory behaviors based on gender, religion, or other identity factors (Younus and Qureshi, 2022; Abburi et al., 2021; Frenda et al., 2019). Given the se- riousness of the problem, researchers in natural language processing (NLP) have been inspired to develop novel methods for sexism in text detection and classification (Butt et al., 2021). This paper discusses the JUST_ONE team’s con- tribution to SemEval 2023-task10. The task aimed to detect and identify various types of sexist con- tent using Gab and Reddit data (Kirk et al., 2023). We used the pre-trained language models: BERT (Devlin et al., 2018), and RoBERTa / unlabeled RoBERTa (Liu et al., 2019) from Hugging face, alongside the ensemble method (selective ensem- ble), and optimized the models’ hyperparameters for each of the three subtasks (A, B, C). It is worth mentioning that our team ranked 19th out of 84 in task A, with a Macro-F1 score of 0.8538, 22nd out …

Content truncated to 500 words