iREL at SemEval-2023 Task 10: Multi-level Training for Explainable Detection of Online Sexism

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 1691–1696 July 13-14, 2023 ©2023 Association for Computational Linguistics iREL at SemEval-2023 Task 10: Multi-level Training for Explainable Detection of Online Sexism Nirmal Manoj, Sagar Joshi, Ankita Maity, Vasudeva Varma IIIT Hyderabad, India {nirmal.manoj, sagar.joshi, ankita.maity}@research.iiit.ac.in vv@iiit.ac.in Abstract This paper describes our approach for SemEval- 2023 Task 10: Explainable Detection of Online Sexism (EDOS). The task deals with identifi- cation and categorization of sexist content into fine-grained categories for explainability in sex- ism classification. The explainable categoriza- tion is proposed through a set of three hierar- chical tasks that constitute a taxonomy of sexist content, each task being more granular than the former for categorization of the content. Our team (iREL) participated in all three hierarchi- cal subtasks. Considering the inter-connected task structure, we study multilevel training to study the transfer learning from coarser to finer tasks. Our experiments based on pretrained transformer architectures also make use of ad- ditional strategies such as domain-adaptive pre- training to adapt our models to the nature of the content dealt with, and use of the focal loss objective for handling class imbalances. Our best-performing systems on the three tasks achieve macro-F1 scores of 85.93, 69.96 and 54.62 on their respective validation sets. 1 Introduction The proliferation of online sexism targeted towards women not only makes online spaces inhospitable but also perpetuates existing social inequities. This has lead to the use of online systems for the identifi- cation of such content through automated technolo- gies based on NLP methods. However, a blatant categorization into general, high-level categories as is often performed by such technologies leads to lack of explainability and trust in these systems. The nature of sexist content being varied based on the motivation of the writer, a discrete identification of categories as different as threats and prejudiced expressions is beneficial for differential treatment. A granular study also helps in identifying the weak- nesses of the model in identifying certain specific categories of sexist content. In our work, we work on the dataset for Explain- able Detection of Online Sexism (introduced as a part of the SemEval-2023 Task 10) (Kirk et al., 2023) based on English language posts from the social media platforms Reddit1 and Gab2. The dataset, discussed in a greater detail in Section 3, introduces three inter-connected subtasks (labeled A, B, C) for the problem, each dealing with classi- fication of sexist comments in a more fine-grained sense. Task A consists of a simple binary classifi- cation of posts into sexist and non-sexist categories. For posts which are detected as sexist, Task B deals with their categorization into 4 distinct categories of sexism. Finally, in Task C looks at a more granu- lar classification of the sexist posts into 11 classes, or “fine-grained vectors of sexism”. We worked on all the three subtasks, and report our experimental findings on the same. To benefit from the inter-connected, hierarchi- cal nature of the three subtasks forming a taxon- omy of sexist …

Content truncated to 500 words