JCT_DM at SemEval-2023 Task 10: Detection of Online Sexism: from Classical Models to Transformers

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 739–743 July 13-14, 2023 ©2023 Association for Computational Linguistics JCT_DM at SemEval-2023 Task 10: Detection of Online Sexism: from Classical Models to Transformers Luzzon Efrat Liebeskind Chaya Department of computer science, Jerusalem College of Technology 21 Havaad Haleumi St., P.O.B 16031 9116991 Jerusalem, Israel {e.luzzon6, liebchaya} @gmail.com Abstract This paper presents the experimentation of sys- tems for detecting online sexism relying on classical models, deep learning models, and transformer-based models. The systems aim to provide a comprehensive approach to han- dling the intricacies of online language, in- cluding slang and neologisms. The dataset consists of labeled and unlabeled data from Gab and Reddit, which allows for the develop- ment of unsupervised or semi-supervised mod- els. The system utilizes TF-IDF with classical models, bidirectional models with embedding, and pre-trained transformer models. The paper discusses the experimental setup and results, demonstrating the effectiveness of the system in detecting online sexism. 1 Introduction Sexism is a pervasive problem online, particularly for women who are disproportionately targeted. It can cause harm, make online spaces unwelcom- ing, and reinforce social inequalities. Automated tools have been developed to identify sexist con- tent at scale, but most current systems only provide generic, high-level classifications without further explanation. This lack of detail makes it difficult for users and moderators to understand and trust the decisions made by these systems. SemEval 2023 - Task 10 - Explainable Detection of Online Sexism (EDOS) (Kirk et al., 2023) aimed at releasing a dataset to assist in the development of English-language models for sexism detection that are more accurate and explainable. This paper handles the first subtask A in task 10. We experi- mented with systems relying on classical models, deep-learning models, and transformer-based mod- els to tackle the first task of binary sexism detection on Gab and Reddit. The goal of our system is to provide a comprehensive approach that can han- dle the intricacies of online language, which often involves slang, colloquialisms, and neologisms. 2 Related Work Previous competition "sEXism Identification in So- cial neTworks (EXIST)" (Rodríguez-Sánchez et al., 2021) was conducted during the IberLEF 2021 in- ternational forum, which was exclusively focused on the field of Natural Language Processing. The competition aimed to classify multilingual posts into five categories, including ideological and in- equality, stereotyping and dominance, objectifica- tion, sexual violence, misogyny, and non-sexual violence. The SemEval competition differs and in- volves the identification of 11 types of sexual posts, indicating a more complex and nuanced classifica- tion task. In addition, Fortuna (Fortuna et al., 2022) ad- dresses the challenges that arise when applying conventional NLP methodologies to hate speech de- tection. The authors reflect on the inadequacies of frequently used conventions for text classification with respect to hate speech detection and discuss the future of current methodologies for this task. Based on their analysis, they conclude that cur- rent models are incapable of detecting hate speech without causing harm to marginalized communities. Therefore, they call for …

Content truncated to 500 words