JudithJeyafreeda at SemEval-2023 Task 10: Machine Learning for Explainable Detection of Online Sexism

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1324–1328 July 13-14, 2023 ©2023 Association for Computational Linguistics JudithJeyafreeda at SemEval-2023 Task 10: Machine Learning for Explainable Detection of Online Sexism Judith Jeyafreeda Andrew University of Manchester Abstract The rise of the internet and social media plat- forms has brought about significant changes in how people interact with each another. For a lot of people, the internet have also become the only source of news and information about the world. Thus due to the increase in acces- sibility of information, online sexism has also increased. Efforts should be made to make the internet a safe space for everyone, irrespective of gender, both from a larger social norms per- spective and legal or technical regulations to help alleviate online gender-based violence. As a part of this, this paper explores simple meth- ods that can be easily deployed to automatically detect online sexism in textual statements. 1 Introduction The Internet is not an equal space. Over the past few years there has been a rise in concerns about the disproportionate levels of abuse experi- enced by women in social media platforms. On- line abuse can take different forms including bul- lying, stalking, impersonation, non-consensual pornography, revenge porn or image-based sex- ual abuse/exploitation, and most commonly, hate speech against women or online misogyny ((Sit, a)). A study ((Sit, a)) has shown that women who experience online abuse often adapt their online behaviour, self-censor the content they post and limit interactions on the platform out of fear of vio- lence and abuse However, despite these there still exists a gap to bridge. (Sit, b) explains the need for identifying and stopping online sexism. One of the reason being: Online gender-based violence can have significant psychological, social, and eco- nomic impacts. Most directly, it affects women’s freedom of expression. Thus in this task we aim at automatically identifying online sexism (gen- der based abusive statements) by taking advantage of Machine Learning methods. In particular, this work explores certain machine learning methods to identify and classify sexist statements in texts into predefined categories. This constitutes a multi- class classification problem.3. 2 Related Work Sexism detection can be characterized as hate speech detection. Several works have been done in this area of research [(Yin and Zubiaga, 2021), (Chetty and Alathur, 2018), (Gambäck and Sik- dar, 2017),(Andrew, 2021b)]. There are several ac- counts of sexist content on major platforms such as Twitter, motivating the development of models for better detecting and classifying social media posts. Thus, most works focus on developing methods for identifying and classifying text in social media platforms [(Pamungkas et al., 2020),(Chiril et al., 2020),(Rodríguez-Sánchez et al., 2020)]. Most state-of-the-art methodologies can be summarized with the following methods: (i) methods using lex- icons (ii) Deep Learning methods, which are more generic but lack domain knowledge (iii) a combina- tion of lexicons and deep learning methods, which is a hybrid method. Over the years, several research work have been conducted on this topic …

Content truncated to 500 words