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

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 469–476 July 13-14, 2023 ©2023 Association for Computational Linguistics Coco at SemEval-2023 Task 10: Explainable Detection of Online Sexism Kangshuai Guo♠, Ruipeng Ma ♠, Shichao Luo♣, Yan Wang♠ ∗ ♣China CITIC Baixin Bank Corporation Limited ♠University of Electronic Science and Technology of China {202122080434,202221080135}@std.uestc.edu.cn xiaoluoyfy@gmail.com,yanbo1990@uestc.edu.cn Abstract Sexism is a growing concern on social media platforms because it affects the overall health of the internet and can have negative impacts on society. This paper describes the coco sys- tem that participated in SemEval-2023 Task 10, Explainable Detection of Online Sexism (EDOS), which aims at sexism detection in var- ious settings of natural language understanding. We develop a novel neural framework for sex- ism detection and misogyny that can combine text representations obtained using pre-trained language model models such as Bidirectional Encoder Representations from Transformers and using BiLSTM architecture to obtain the local and global semantic information. Further, considering that the EDOS dataset is relatively small and extremely unbalanced, we conducted data augmentation and introduced two datasets in the field of sexism detection. Moreover, we introduced Focal Loss which is a loss function in order to improve the performance of process- ing imbalanced data classification. Our system achieved an F1 score of 78.95% on Task A - binary sexism. 1 Introduction Social media platforms offer unparalleled commu- nication and information-sharing abilities while providing an anonymous space for people of vary- ing genders, ethnicities, races, and cultures to inter- act online. (Fortuna et al., 2021; Pamungkas et al., 2021; Chiril et al., 2020; Williams et al., 2020). As a result, there has been a rise in incidents, hostile behavior, and instances of harassment on social me- dia platforms, especially sexism (Kirk et al., 2023; Deluca et al., 2016). Sexism is connected to be- liefs about the essential nature of women and men, as well as the roles they should assume in society. This hierarchical way of thinking can be deliberate and hostile, or it can be unintentional and take the form of unconscious bias. ∗: Corresponding author. The SemEval task 10 supports the development of English-language models for sexism detection that fine-grained classifications for sexist content from Gab and Reddit, which consists of three sub- tasks (Kirk et al., 2023). As illustrated in Fig.1: (1) Task A - Binary Sexism Detection: It is a binary classification task which models are required to predict whether a text contains gender bias or not; (2) Task B - Category of Sexism: It is a four-class classification task on texts that exhibit gender bias which requires models to predict one of four categories: (1) threats, (2) derogation, (3) animosity, (4) prejudiced discussions; (3) Task C - Fine-grained Vector of Sexism : It is an 11-class classification task for texts ex- hibiting gender bias which requires models to predict one of 11 fine-grained vectors. Our method employed Bert (Devlin et al., 2018) as mono-lingual pre-trained language models, and various models including CNN, LSTM and BiL- STM …

Content truncated to 500 words