SUTNLP at SemEval-2023 Task 10: RLAT-Transformer for explainable online sexism detection

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 347–356 July 13-14, 2023 ©2023 Association for Computational Linguistics SUTNLP at SemEval-2023 Task 10: RLAT-Transformer for explainable online sexism detection Hamed Hematian Hemati, Sayed Hesam Alavian, Hamid Beigy, Hossein Sameti AI Group, Computer Engineering Department, Sharif University of Technology, Tehran, Iran { hematian.hemati.hamed, hesam.alavian70}@gmail.com, {beigy, sameti}@sharif.edu Abstract There is no simple definition of sexism, but it can be described as prejudice, stereotyping, or discrimination, especially against women, based on their gender. In online interactions, sexism is relatively rare but still harmful. One out of ten American adults says that they have been harassed because of their gender and have been the target of sexism, so sexism is a growing issue. The Explainable Detection of Online Sexism shared task in SemEval-2023 aims at building sexism detection systems for the English language. In order to address the problem, we use large language models such as RoBERTa and DeBERTa. In addition, we present a novel method called Random Layer Adversarial Training (RLAT) for transform- ers which is based on adversarial training, and show its impact on boosting all subtasks’ scores. Moreover, we use other discriminative and gen- eralization techniques for subtask A to boost performance. Using our methods to make pre- dictions over subtask A, B, and C test sets, we obtained macro-F1 of 84.45, 67.78, and 52.52 respectively, outperforming proposed baselines on all subtasks. Our code is publicly available on Github.1 1 Introduction Sexism refers to any negative, abusive, or dis- criminatory behavior, attitude, or language that targets individuals based on their gender, particu- larly women, and involves prejudice, stereotyping, or discrimination (Kirk et al., 2023). In essence, sexism encompasses any negative, abusive, or dis- criminatory behavior or attitude that targets women based on their gender or a combination of their gender with other identity characteristics such as race, religion, or gender identity, for instance, black women, Muslim women, or trans women (Kirk et al., 2023). The internet is a breeding ground 1https://github.com/SUTNLP/RLAT-Transformer for sexism, with approximately one in ten Ameri- can adults reporting that they have been harassed due to their gender, and have experienced sexist behavior (Swim et al., 2001). In order to promote a fair and inclusive online environment, it is vital to acknowledge and com- bat sexism on online platforms because of their significant scale, extensive reach, and powerful in- fluence.2 Sexism poses a growing problem in the online environment, and its consequences can be severe, such as making online spaces inaccessible and per- petuating social injustices. To address this issue, various automated tools have been deployed to de- tect and assess sexist content. Nonetheless, most of these tools only provide generic classifications without further explanation. Identifying sexist con- tent and explaining why it is offensive will increase an automated tool’s interpretability, trustworthi- ness, and understanding of its decisions. This, in turn, will empower both users and moderators to make informed decisions and create a fair and in- clusive online environment (Kirk et al., 2023). The …

Content truncated to 500 words