CSECU-DSG at SemEval-2023 Task 10: Exploiting Transformers with Stacked LSTM for the Explainable Detection of Online Sexism

Proceedings of the The 17th International Workshop on Semantic Evaluation (SemEval-2023), pages 2091–2099 July 13-14, 2023 ©2023 Association for Computational Linguistics CSECU-DSG at SemEval-2023 Task 10: Exploiting Transformers with Stacked LSTM for the Explainable Detection of Online Sexism Afrin Sultana, Radiathun Tasnia, Nabila Ayman and Abu Nowshed Chy Department of Computer Science and Engineering University of Chittagong, Chattogram-4331, Bangladesh afrin.sultana.cu@gmail.com, radia.tasnia.cu@gmail.com, nabila.ayman.cu@gmail.com, and nowshed@cu.ac.bd Abstract Sexism is a harmful phenomenon that provokes gender inequalities and social imbalances. The expanding application of sexist content on so- cial media platforms creates an unwelcoming and discomforting environment for many users. The implication of sexism is a multi-faceted subject as it can be integrated with other cat- egories of discrimination. Binary classifica- tion tools are frequently employed to identify sexist content, but most of them provide ex- tensive, generic categories with no further in- sights. SemEval-2023 introduced the Explain- able Detection of Online Sexism (EDOS) task that emphasizes detecting and explaining the category of sexist content. The content of this paper details our involvement in this task where we present a neural network architecture employing document embeddings from a fine- tuned transformer-based model into stacked long short-term memory (LSTM) and a fully connected linear (FCL) layer . Our proposed methodology obtained an F1 score of 0.8218 (ranked 51st) in Task A. It achieved an F1 score of 0.5986 (ranked 40th) and 0.4419 (ranked 28th) in Tasks B and C, respectively. Keywords: Sexism . Transformer document em- beddings . Stacked LSTM . BERTweet . Fully connected layer Content Warning: This manuscript discusses examples of hateful and sexist content. The au- thors do not support the use of hateful and sexist language, nor any of the hateful and sexist repre- sentations quoted. 1 Introduction The ongoing progress and expansion of technol- ogy and social platforms are driving the growth of user-generated content. Online platforms (e.g., Facebook, Twitter, blogs, and forums) reflect a wide range of perspectives, facts, and experiences The first three authors have equal contributions. from a diverse set of people. In addition, social media, which offers an anonymous atmosphere, also enables users to display hostile sentiments that can result in harassment of individuals with vari- ous genders, nations, races, cultures, and physical features (Fox et al., 2015; Litchfield et al., 2018). Online sexism is a rising issue nowadays. Sexism is a prevalent form of hate speech that is presently seen as a negative trend on social media. Sexism involves the unjust and discriminatory treatment of women or holding negative attitudes towards them solely because of their gender or because of their gender in combination with other aspects of their identity, such as their race, religion. (Bad- jatiya et al., 2017; Sharifirad and Matwin, 2019; Kirk et al., 2023). Automatic approaches (Badjatiya et al., 2017; Davidson et al., 2017; Del Vigna12 et al., 2017; Fersini et al., 2018; Basile et al., 2019) are increas- ingly used to find and assess sexist content at scale, however, most of them just classify information into general, high-level categories …

Content truncated to 500 words