AU_NLP at SemEval-2023 Task 10: Explainable Detection of Online Sexism Using Fine-tuned RoBERTa

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 707–717 July 13-14, 2023 ©2023 Association for Computational Linguistics AU_NLP at SemEval-2023 Task 10: Explainable Detection of Online Sexism Using Fine-tuned RoBERTa Amit Das1, Nilanjana Raychawdhary1, Tathagata Bhattacharya2, Gerry Dozier1, and Cheryl D. Seals1 1Department of Computer Science & Software Engineering, Auburn University, AL, USA 2Department of Computer Science, Auburn University at Montgomery, AL, USA 1{azd0123,nzr0044,doziegv,sealscd}@auburn.edu 2{tbhatta1}@aum.edu Abstract Social media is a concept developed to link people and make the globe smaller. But it has recently developed into a center for sexist posts that target especially women. As a result, there are more events of hostile actions and harassing remarks present online. In this paper, we in- troduce our system for the task of Explainable Detection of Online Sexism (EDOS), a part of SemEval 2023 task 10. We introduce a fine- tuned RoBERTa model by encoding the initial representation of the data using Roberta and setting three distinct Multilayer Perceptrons (MLPs) corresponding to the three sub-tasks to address this specific problem. The effective- ness of the proposed strategy is demonstrated by the experimental results reported in this re- search. 1 Introduction Discriminatory ideas and sentiments, particularly those directed at women, are frequently found on- line and on social media. This typically indicates the presence of additional hazardous content types like hate speech (Demus et al., 2022) or misinfor- mation (Schütz et al., 2021). It might be difficult to identify such remarks because sexism and misog- yny can take many different forms and vary across linguistic and cultural boundaries. Sexism is one of the most critical difficulties because it may greatly harm users. As an illus- tration, objectification, a kind of sexism, has been linked to serious risks for eating disorders, unipolar depression, and sexual dysfunction (Fredrickson and Roberts, 1997). A study published by Fox et al. (Fox et al., 2015) requested participants to retweet or post tweets with sexist content before conducting a set of activities meant to uncover sex- ist behaviors. Additionally, since the anonymity of Twitter can encourage people to demonstrate an even greater sexist behavior, the authors came to the conclusion that the users who were protected by the anonymity of a social media profile were more likely than non-anonymous users to exhibit aggressive sexism. Sexism is defined by the Oxford English Dictio- nary as prejudice, stereotyping or discrimination, typically against women, on the basis of sex.1 Sex- ist attitudes and languages undervalue the contri- bution of women. Furthermore, considering that a sizeable portion of Internet users, particularly those who utilize social networks, are teenagers, the rise in online sexism necessitates urgent research and social debate that results in action (Rodríguez- Sánchez et al., 2020). However, detecting online sexism may be diffi- cult, as it may be expressed in very different forms. In this study, we focus mainly on using Natural Language Processing (NLP) methods and state-of- the-art models for two important tasks: (i) binary sexism detection, which aims to determine whether a given sentence …

Content truncated to 500 words