UL & UM6P at SemEval-2023 Task 10: Semi-Supervised Multi-task Learning for Explainable Detection of Online Sexism

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 644–650 July 13-14, 2023 ©2023 Association for Computational Linguistics UL & UM6P at SemEval-2023 Task 10: Semi-Supervised Multi-task Learning for Explainable Detection of Online Sexism Salima Lamsiyah1, Abdelkader El Mahdaouy2, Hamza Alami3 Ismail Berrada3 and Christoph Schommer1 1Dept. of Computer Science, Faculty of Science, Technology and Medicine, University of Luxembourg, Luxembourg 2Modeling, Simulation and Data Analysis, Mohammed VI Polytechnic University, Morocco 3School of Computer Science, Mohammed VI Polytechnic University, Morocco {firstname.lastname}@{uni.lu1, um6p.ma2,3} Abstract This paper introduces our participating system to the Explainable Detection of Online Sex- ism (EDOS) SemEval-2023 - Task 10: Explain- able Detection of Online Sexism. The EDOS shared task covers three hierarchical sub-tasks for sexism detection, coarse-grained and fine- grained categorization. We have investigated both single-task and multi-task learning based on RoBERTa transformer-based language mod- els. For improving the results, we have per- formed further pre-training of RoBERTa on the provided unlabeled data. Besides, we have em- ployed a small sample of the unlabeled data for semi-supervised learning using the mini- mum class-confusion loss. Our system has achieved macro F1 scores of 82.25%, 67.35%, and 49.8% on Tasks A, B, and C, respectively. 1 Introduction The advent of diverse social media platforms has enabled both identified and anonymous users to express their views and beliefs freely and openly. However, a number of online users leverage this freedom of expression to abuse other people or a group of people based on their race, ethnicity, culture, religion, gender, and political orientation, to name but a few (Guiora and Park, 2017). To tackle this issue, social media platforms rely on content moderation techniques to review and fil- ter their users’ posts (Shen and Rose, 2019; Gille- spie, 2018, 2020; Liu et al., 2021). During the past decade, researchers have shown an increased inter- est in building artificial intelligence-based tools and applications to address the challenges of content moderation automation in social media platforms. For instance, various Natural Language Processing (NLP) research works have been introduced for of- fensive language (Zampieri et al., 2019, 2020), hate speech (Basile et al., 2019; Röttger et al., 2021), Cyberbullying (Van Hee et al., 2015; Menini et al., 2019; Chen and Li, 2020), toxicity (van Aken et al., 2018; Pavlopoulos et al., 2021), misogyny (Fersini et al., 2018; Mulki and Ghanem, 2021) and sexism (Jha and Mamidi, 2017; Rodríguez-Sánchez et al., 2021) detection. The Explainable Detection of Online Sexism (EDOS) shared task introduces three sub-tasks for sexism detection and categorization in the English language (Kirk et al., 2023). The aim is to develop NLP models and systems for sexism detection and explainability. It emphasizes improving predic- tions’ interpretability and explainability for fair content moderation decision-making. The EDOS shared task provides training data for developing an accurate and explainable model for sexism de- tection using a set of hierarchical labels and tasks. In this paper, we present our participating system to the EDOS shared task. We have explored both single-task and multi-task learning models. …

Content truncated to 500 words