I2C-Huelva at SemEval-2023 Task 10: Ensembling Transformers Models for the Detection of Online Sexism

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 763–769 July 13-14, 2023 ©2023 Association for Computational Linguistics 1 Abstract This work details our approach for addressing Tasks A and B of the Semeval 2023 Task 10: Explainable Detection of Online Sexism (EDOS). For Task A a simple ensemble based of majority vote system was presented. To build our proposal, first a review of transformers was carried out and the 3 best performing models were selected to be part of the ensemble. Next, for these models, the best hyperpameters were searched using a reduced data set. Finally, we trained these models using more data. During the development phase, our ensemble system achieved an f1-score of 0.8403. For task B, we developed a model based on the deBERTa transformer, utilizing the hyperparameters identified for task A. During the development phase, our proposed model attained an f1-score of 0.6467. Overall, our methodology demonstrates an effective approach to the tasks, leveraging advanced machine learning techniques and hyperparameters searches to achieve high performance in detecting and classifying instances of sexism in online text. 1 Introduction Sexism refers to any type of mistreatment or negative attitude that is specifically directed towards women based on their gender or based on their gender combined with one or more additional aspects of their identity (such as being a black woman, Muslim woman, or transgender woman). Allowing sexism to thrive on these platforms can normalize violence against women and other individuals who face discrimination based on gender or gender identity. Therefore, it is crucial to take measures to eradicate sexism on social media and promote a more inclusive and respectful environment for all individuals. In this work, were used Transformer- based solutions (Wolf et al. 2020) and follow an ensemble strategy, specifically for the sexism detection in Task A. By completing these tasks, we became aware selecting the correct hyperparameters for each model also boosted the prediction rate for the model. 2 Background A dataset developed for the SemEval 2023 Task - Explainable Detection of Online Sexism (Kirk et al. 2023) is used to train, validate, and test models. This dataset contains labelled data from the social network Twitter in English language. Labels associated to each tweet gives information used to classify if the tweet is sexist or not (for Task A), and in case that the tweet is sexist, categorize the sexism of the tweet (for Task B). The total size of the dataset is 13984 tweets (10602 not sexist and 3398 sexist). The dataset, common to all tasks, contains the tweets along with the labels used for their classification. These labels are divided in two types: sexist label (used to detect if the tweet is sexist or not sexist) and category label (used to categorize the sexism of the tweet threats, derogation, animosity, and prejudiced discussion). For Task A, based on the strategy described in (Vaca-Serrano 2022) for a similar task, a simple ensemble based of majority vote system was presented. First a review …

Content truncated to 500 words