KingsmanTrio at SemEval-2023 Task 10: Analyzing the Effectiveness of Transfer Learning Models for Explainable Online Sexism Detection

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1916–1920 July 13-14, 2023 ©2023 Association for Computational Linguistics KingsmanTrio at SemEval-2023 Task 10: Analyzing the Effectiveness of Transfer Learning Models for Explainable Online Sexism Detection Fareen Tasneem, Tashin Hossain, and Jannatun Naim Department of Computer Science and Engineering University of Chittagong, Chattogram-4331, Bangladesh {fareen.tasneem, tashin.hossain.cu, and jannatun.naim.cu}@gmail.com Abstract Online social platforms are now propagating sexist content endangering the involvement and inclusion of women on these platforms. Sex- ism refers to hostility, bigotry, or discrimina- tion based on gender, typically against women. The proliferation of such notions deters women from engaging in social media spontaneously. Hence, detecting sexist content is critical to en- sure a safe online platform where women can participate without the fear of being a target of sexism. This paper describes our participation in subtask A of SemEval-2023 Task 10: Ex- plainable Detection of Online Sexism (EDOS). This subtask requires classifying textual con- tent as sexist or not sexist. We incorporate a RoBERTa-based architecture and further fine- tune the hyperparameters to entail better perfor- mance. The procured results depict the compet- itive performance of our approach among the other participants. 1 Introduction The free flow of information is the most fascinating feature of social media that has enabled sharing of information, thoughts, and opinions across the world. But this very feature is now used as a tool to incite hatred and threats against individuals or com- munities. Our society has always exploited women with prejudice, stereotype, and discrimination. The creation and proliferation of social media have pro- vided our society with a more extensive platform to induce such unjust treatment against women. Peo- ple can post sexist content anonymously using fake and untraceable identities without the fear of ac- countability. Moreover, the hatred can be spread to the wider community impeding the integration of women into the social network. So it is essential to detect such sexism in social media content so that women can partake and collaborate in social me- dia without being mistreated and threatened. With All of the authors have equal contributions. this intention, SemEval-2023 Task 10: Explainable Detection of Online Sexism (EDOS) (Kirk et al., 2023) is introduced. The aim of the task is to detect sexism in the textual content of social media. The task is divided into three hierarchical tasks. We participated in the subtask A - Binary Sexism De- tection where a text must be classified as sexist or not-sexist. To explicate this task, two examples are shown in 1. Sample texts Class #1: Im beginning to see why women werent allowed in poli- tics or allowed to vote. Sexist #2: Kaine is so creepy He looks like a raging alcoholic wife beater Ick Not-sexist Table 1: Sexist and Non-sexist samples. Here, in the first example, the capability of women in political activities is being doubted just based on their sex. Hence, the first text is consid- ered sexist. On the other hand, the second example depicts violence …

Content truncated to 500 words