PoSh at SemEval-2023 Task 10: Explainable Detection of Online Sexism

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1276–1281 July 13-14, 2023 ©2023 Association for Computational Linguistics PoSh at SemEval-2023 Task 10: Explainable Detection of Online Sexism Shruti Sriram, Padma Pooja Chandran and Shrijith M R Department of Computer Science and Engineering Sri Sivasubramaniya Nadar College of Engineering, Chennai, India shruti2010197@ssn.edu.in, padmapooja2010291@ssn.edu.in, shrijith2010542@ssn.edu.in Abstract The growing popularity of online platforms and social media has intensified the problem of sex- ism, making it more prevalent and widespread. Online platforms have provided an anonymous space where individuals can freely express their misogynistic beliefs with discriminatory and abusive language, creating a culture of online harassment and hate speech, mainly targeting women. The dataset used in this task was pro- vided by SemEval-2023 Task 10: Explainable Detection of Online Sexism. To precisely iden- tify the different forms of online sexism, we uti- lize several sentence transformer models such as ALBERT, BERT, RoBERTa, DistilBERT, and XLNet. By combining the predictions from these models, we can generate a more compre- hensive and improved result. Each transformer model is trained after pre-processing the data from the training dataset. Our team obtained macro f1 scores of 0.7937 for subtask A, 0.5284 for subtask B and 0.2674 for subtask C. 1 Introduction In today’s technology-driven world, it has become increasingly common for users to utilize social media platforms to spread hatred and engage in sexist behaviour. Therefore, addressing online sexism is a critical way to create a safe online space for everyone, where people can express their opinions without fear of discrimination or harass- ment. Online sexism is a form of gender-based discrimination or harassment that occurs in online spaces, including social media platforms, online forums, chat rooms, and messaging apps. It can take many forms, from overly sexist comments, hashtags (Fox et al., 2015) and slurs in the forms of discrimination, such as dehumanizing language, gender-based stereotypes, and microaggressions. Further, it can have a significant impact on the mental and emotional status of individuals who experience it. Also, it can lead to feelings of shame, anxiety, and depression, as well as physical symptoms such as headaches and insomnia. Online sexism can be directed at individuals or groups of people based on their gender, sexual orientation, race, religion, or other personal characteristics. Women are particularly vulnerable to online sexism as they are more likely to be targeted with derogatory comments and threats of violence, and overall experience higher levels of online harassment. Given the rising issue of online sexism, it is essential to develop effective strategies for detecting and addressing the same. One promising approach is to use natural language processing (NLP) transformers to analyze social media data and identify instances of online sexism. We have participated and submitted our solu- tions for all three subtasks. The goal of these tasks is to enhance the accuracy and explainability of English-language models for detecting online sexism. The main focus is to create fine-grained classifications for sexist content on Gab and Reddit (based on training data). By …

Content truncated to 500 words