Contextual Cyberbullying Detection using DeBERTa and LSTM for Sequential Text Analysis

Cyberbullying has been on the rise over the years due to the growing popularity of social media platforms and communication tools, which pose serious emotional and psychological hazards to users, especially the younger generation. Although various machine learning and deep learning approaches have been proposed for the detection of such toxic content, traditional methods like SVMs or basic neural networks are surface-level approaches and cannot possibly tap into the implicit, context-driven, and subtle nature of abusive language. Even transformer-based approaches like BERT, which are inherently context-driven, may not be able to grasp the subtle nuances of abuse. To overcome these limitations, we propose a hybrid approach to cyberbullying detection that combines the context understanding capabilities of DeBERTa with the ability of LSTM to process sequences. In our approach, DeBERTa is used to generate deep, semantic representations of the input text, which are then further processed by LSTM to identify any sequential patterns and relationships that may exist and potentially indicate cyberbullying. The approach ensures end-to-end detection of cyberbullying and provides a balanced, interpretable, and accurate solution that harnesses the power of deep learning and advanced language modeling.