Leveraging DistilBERT and BERT for the Detection of Online Sexism: A Comparative Analysis
Sowmen Mitra, Proma Kanungoe, Ali Newaz Chowdhury · 2023
Online sexism perpetuates harmful gender stereotypes and biases, leading to an environment rife with prejudice and injustice. This not only erodes human well-being by inducing feelings of emotional distress and worthlessness, particularly among women and marginalized genders, but also stifles the free exchange of ideas by creating hostile digital spaces. These spaces suppress voices, limit participation, and hinder meaningful interactions. In our study, we utilized the Bidirectional Encoder Representations from Transformers (BERT) and DistilBERT to swiftly identify sexist comments. Experimental results indicate that BERT significantly outperforms both DistilBERT and other leading architectures. With an emphasis on sustainable AI practices, we've optimized both models for efficiency. In evaluating their efficacy, we've considered both their F1 score and their environmental impact.