| Resumen |
In this paper, we address the Task 1 and Task 2 of the EXIST 2022 in detecting sexism in a broad sense, from ideological inequality, sexual violence, misogyny to other expressions that involve implicit sexist behaviours in social networks. We apply transfer learning from a pre-trained multilingual DeBERTa (mDeBERTa) model and its zero classification to gain a better performance than BERT-based approaches. Lastly, we combine all 3 methods: mDeBERTa, zero classification, and BERT for majority vote. For Task 1, mDeBERTa is the best method with an accuracy of 76.09% and F1 of 76.08%. Meanwhile, an accuracy of 66.26% and F1 of 47.06% are the best results in Task2, when using majority vote. Our main contribution is to use DeBERTa and zero classification with designing only one classifier in sexism identification. © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). |
