Transformer Models for Combating Cyberbullying: Comparative Evaluation on Racism and Sexism Detection
Students & Supervisors
Student Authors
Supervisors
Abstract
Racism and sexism’s constitute the majority of online abuse and still remains one of the main instigator for cyberbullying on the internet. Online bullying inflict severe mental harm and also affect the safety of women expressing their views online which otherwise always considered welcoming in nature. This research tries to investigates the current trans former based state of the art NLP models and their capabilities toward identifying racist and sexist conversation in social media discourse. More specifically, BERT, XLNet, and RoBERTa are compared against each other on multiple datasets accumulated from social media websites. Following the normalization and balancing operations, the final datasets had 3,940 samples related to racism and 6,746 samples related to sexism. Each transformer model was fine tuned and assessed based on these performance indicators: accuracy, recall, precision, F1 score, and AUC-ROC. The findings revealed that, in all evaluated datasets, RoBERTa topped BERT and XLNet. RoBERTa in the racism dataset achieved 91.88% accuracy and an AUC of 0.959, also acquired an AUC-ROC of 0.906 and 83.11% accuracy on the sexism dataset. This shows it’s outstanding ability to recognize and capture abusive words in subtle and context sensitive uses of the given literature. Insights gathered from this work may drive future investigations and the help us find feasible solutions to this issue of cyberbullying in virtual environment.
Keywords
Publication Details
- Type of Publication:
- Conference Name: 11th IEEE International Women in Engineering (WIE) Conference on Electrical and Computer Engineering 2025
- Date of Conference: 21/12/2025 - 21/12/2025
- Venue: Long Beach Hotel, Cox’s Bazar, Bangladesh
- Organizer: IEEE Bangladesh Section and IEEE Women in Engineering (WIE)