DUTIR at SemEval-2023 Task 10: Semi-supervised Learning for Sexism Detection in English

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 892–896 July 13-14, 2023 ©2023 Association for Computational Linguistics DUTIR at SemEval-2023 Task 10: Semi-supervised Learning for Sexism Detection in English Bingjie Yu, Zewen Bai, Haoran Ji, Shiyi Li, Hao Zhang, Hongfei Lin* School of Computer Science and Technology, Dalian University of Technology, China (yubingjie,986427968,18920392258,lishiyieee,zh373911345)@mail.dlut.edu.cn hflin@dlut.edu.cn Abstract Sexism is an injustice afflicting women and has become a common form of oppression in social media. In recent years, the automatic detection of sexist instances has been utilized to combat this oppression. The Subtask A of SemEval- 2023 Task 10, Explainable Detection of On- line Sexism, aims to detect whether an English- language post is sexist. In this paper, we de- scribe our system for the competition. The structure of the classification model is based on RoBERTa, and we further pre-train it on the domain corpus. For fine-tuning, we adopt Un- supervised Data Augmentation (UDA), a semi- supervised learning approach, to improve the robustness of the system. Specifically, we em- ploy Easy Data Augmentation (EDA) method as the noising operation for consistency train- ing. We train multiple models based on dif- ferent hyperparameter settings and adopt the majority voting method to predict the labels of test entries. Our proposed system achieves a Macro-F1 score of 0.8352 and a ranking of 41/84 on the leaderboard of Subtask A. 1 Introduction With the evolution of the internet, people are free to express their opinions on social media. This may lead to the mass dissemination of hateful or abusive messages (Chiril et al., 2020), such as sex- ist expressions that people use intentionally or un- intentionally. Sexism is a complex phenomenon, broadly defined as "prejudice, stereotyping, or dis- crimination, typically against women, on the basis of sex."1 Sexism takes many forms, including bla- tant, covert, and subtle sexism (Swim et al., 2004), causing sexism detection to be a challenge for the filtering mechanism of platforms. The SemEval-2023 Task 10 (Kirk et al., 2023) is an explainable detection task of online sexism, and we participate in Subtask A, the binary classifi- cation task, to detect whether an English-language 1Oxford English Dictionary post is sexist or not. The competition organizers released datasets collected from Gab and Reddit, containing 20,000 labeled entries and 2 million unlabeled entries. For the sexism detection task, we build a system based on the pre-trained language model RoBERTa. Firstly, we further pre-train RoBERTa on the unla- beled corpus to obtain our domain-specific encoder. Furthermore, we reuse the abundant unlabelled data by Unsupervised Data Augmentation (UDA) (Xie et al., 2020) approach for consistency training, which is a common practice in semi-supervised learning. As for the noising operation, we apply the Easy Data Augmentation (EDA) (Wei and Zou, 2019) method to transform the unlabeled examples into their noised versions. The consistency training can propagate label information from labeled to unlabeled examples to enhance the generalization of the model. To further improve the robustness of our system, we train models based on different hy- …

Content truncated to 500 words