NLP_CHRISTINE at SemEval-2023 Task 10: Utilizing Transformer Contextual Representations and Ensemble Learning for Sexism Detection on Social Media Texts

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 595–602 July 13-14, 2023 ©2023 Association for Computational Linguistics NLP_CHRISTINE at SemEval-2023 Task 10: Utilizing Transformer Contextual Representations and Ensemble Learning for Sexism Detection on Social Media Texts Christina Christodoulou Department of Informatics and Telecommunications, National and Kapodistrian University of Athens Institute for Language and Speech Processing, Athena Research Center christinachristodoulou1997@gmail.com Abstract The paper describes the SemEval-2023 Task 10: Explainable Detection of Online Sexism (EDOS), which investigates the detection of sexism on two social media sites, Gab and Reddit, by encouraging the development of machine learning models that perform binary and multi-class classification on English texts. The EDOS Task consisted of three hierarchical sub-tasks: binary sexism detection in sub-task A, category of sexism detection in sub-task B and fine-grained vector of sexism detection in sub-task C. My participation in EDOS com- prised fine-tuning of different layer represen- tations of Transformer-based pre-trained lan- guage models, namely BERT, AlBERT and RoBERTa, and ensemble learning via major- ity voting of the best performing models. De- spite the low rank mainly due to a submis- sion error, the system employed the largest version of the aforementioned Transformer models (BERT-Large, ALBERT-XXLarge-v1, ALBERT-XXLarge-v2, RoBERTa-Large), ex- perimented with their multi-layer structure and aggregated their predictions so as to get the final result. My predictions on the test sets achieved 82.88%, 63.77% and 43.08% Macro-F1 score in sub-tasks A, B and C respectively. 1 Introduction The Task and Its Importance Sexism emerged as a term from the second-wave feminism of the 1960s through the 1980s to comprise discrimination, prejudice and stereotyping based on someone’s sex, typically targeting women and girls (Masequesmay, 2022). Even though our way of life has improved since then, it appears that sexism persists. On top of that, considering the immense influence of social networks on our lives, sexism is expressed online through text, image, video and sound on a daily ba- sis. It has become evident that the anonymity and in- visibility of the online environment foster the online disinhibition effect, meaning the tendency to exhibit negative behavior online rather than in person with- out taking into account the consequences (Wright et al., 2019),(Fox et al., 2015). In recent years, the amount of sexist content in social media ranging from cases of discrimination, abuse, stalking, bully- ing, impersonation, defamation, verbal and sexual harassment, non-consensual pornography, misog- yny to cases of hate speech mainly against female individuals or groups has been on the rise. This has serious repercussions for the the mental health, the social status and the online experience of the targets, not to mention the negative impact on the community. To illustrate, the majority of women, who have experienced or witnessed sexist behavior online, state that they are discouraged from pursu- ing their political careers and upholding their rights, while they are led to self-censor and limit their pres- ence on social media (Meco and MacKay, 2022). What is worse, acts of violence and abuse online can disrupt the targets’ physical …

Content truncated to 500 words