Ebhaam at SemEval-2023 Task 1: A CLIP-Based Approach for Comparing Cross-modality and Unimodality in Visual Word Sense Disambiguation

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 184–192 July 13-14, 2023 ©2023 Association for Computational Linguistics HULAT at SemEval-2023 Task 10: Data Augmentation for Pre-trained Transformers Applied to the Detection of Sexism in Social Media Isabel Segura-Bedmar Computer Science Department, Universidad Carlos III de Madrid, Leganés, Spain isegura@inf.uc3m.es Abstract This paper describes our participation in SemEval-2023 Task 10, whose goal is the de- tection of sexism in social media. We explore some of the most popular transformer models such as BERT, DistilBERT, RoBERTa, and XL- Net. We also study different data augmenta- tion techniques to increase the training dataset. During the development phase, our best results were obtained by using RoBERTa and data aug- mentation for tasks B and C. However, the use of synthetic data does not improve the results for task A. We participated in the three subtasks. Our approach still has much room for improve- ment, especially in the two fine-grained classi- fications. All our code is available in the repos- itory https://github.com/isegura/hulat_ edos. 1 Introduction Sexism can be defined as behaviors or beliefs that support gender inequality, and result in discrimina- tion, generally against women. Contrary to what one might believe, sexism is still very present also in the most advanced and technologically advanced societies (Ridgeway, 2011). Proof of this is that many gender stereotypes are still present in our belief system today (for example, men should not wear dresses). Unfortunately, social networks are used to spread hateful and sexist messages against women (Rodríguez-Sánchez et al., 2020). During the last few years, various research ef- forts (Rodríguez-Sánchez et al., 2022; Fersini et al., 2022) have been devoted to the development of automatic tools for the detection of sexist con- tent. While these automated tools have addressed the classification of sexist content, this is a high- level classification, without providing additional information that allows us to understand why the content is sexist. The goal of SemEval-2023 Task 10, Explainable Detection of Online Sexism (EDOS)(Kirk et al., 2023), is to promote the de- velopment of fine-grained classification models for detecting sexism in posts written in English, which were collected from Gab and Reddit. The organiz- ers of the task proposed three subtasks: A) Binary Sexism Detection, B) Category of Sexism, a four- class classification task, and C) Fine-grained Vector of Sexism, an 11-class classification. A detailed description of these classifications can be found at Kirk et al. (2023). In our approach, we explored some of the most popular pre-trained transformer models such as BERT (Devlin et al., 2019), DistilBERT (Sanh et al., 2019), RoBERTa (Zhuang et al., 2021), and XLNet (Yang et al., 2019). Moreover, we used dif- ferent data augmentation techniques such as EDA (Wei and Zou, 2019) and NLPAug library to create synthetic data.1 Then, synthetic data and training data were used to fine-tune the models. Based on our experiments during the development phase, we decided to use the RoBERTa transformer model to estimate our predictions for the test dataset during …

Content truncated to 500 words