PCJ at SemEval-2023 Task 10: A Ensemble Model Based on Pre-trained Model for Sexism Detection and Classification in English

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 433–438 July 13-14, 2023 ©2023 Association for Computational Linguistics PCJ at SemEval-2023 Task 10: A Ensemble Model Based on Pre-trained Model for Sexism Detection and Classification in English Chujun Pu Yunnan University 1448592081@qq.com Xiaobing Zhou* Yunnan University zhouxb@ynu.edu.cn Abstract This paper describes the system and the result- ing model submitted by our team "PCJ" to the SemEval-2023 Task 10 sub-task A contest. In this task, we need to test the English text con- tent in the posts to determine whether there is sexism, which involves emotional text classifi- cation. Our submission system utilizes meth- ods based on RoBERTa, SimCSE-RoBERTa pre-training models, and model ensemble to classify and train on datasets provided by the organizers. In the final assessment, our sub- mission achieved a macro average F1 score of 0.8449, ranking 28th out of 84 teams in Task A. 1 Introduction Sexism is a very serious problem on the Inter- net nowadays, where unsuspecting people attack women by making statements against them, expos- ing them to unfair treatment, and eventually caus- ing harmful effects on society. Therefore, tools that can automatically detect sexism are widely de- ployed (Bordia and Bowman, 2019), which is an important task in the field of NLP (Mai et al.). Sexism detection is a text classification problem. Text classification is a classical problem in the field of natural language processing, and the main meth- ods can be divided into two categories: machine learning-based methods and deep learning-based methods. For the machine learning-based methods (Agarwal and Mittal), the main idea is artificial fea- ture engineering plus shallow classification model (Li et al., 2020). Text feature engineering is divided into three parts: text preprocessing, feature extrac- tion, and text representation, with the ultimate goal of converting text into a computer-understandable format and encapsulating enough information for classification, and then the classifier uses statis- tical classification methods, such as SVM, plain Bayesian classification algorithms. This approach overall relies heavily on discrete manual features. For deep learning-based text classification (Minaee et al., 2021), the main idea is that the text to be classified is represented as a word vector, and then deep learning networks and other variants are used for automatic feature extraction, represented by models such as Transformer (Vaswani et al., 2017). However, there are still some deficiencies in cur- rent sexism detection, which need to be improved, such as the abuse of language detection models, and the misestimation of sexism (Park et al., 2018). Task 10 of SemEval-2023 is the explainable de- tection of online sexism (Kirk et al., 2023), which aims to address the issue of automated detection of large-scale online sexism. By proposing a hier- archical taxonomy with three tasks for detecting sexist content, it ultimately aims to improve the accuracy and explainability of automated detection tools, and contribute to reducing the harmful effects of online sexism. In this paper, we describe the details of our re- search in Subtask A and how it performed …

Content truncated to 500 words