BiLSTM and BERT-Base-Multilingual-Cased Integration for Cyberbullying Detection: A Multi-Class Approach

This paper outlines a cyberbullying detection method tailored for the Hungarian language, combining BiLSTM and a pretrained BERT-Base Multilingual Cased tokenizer using PyTorch. We collected a Hungarian dataset from online platforms, processed through normalization, tokenization, and encoding. Our evaluation shows the model's effectiveness in identifying cyberbullying in Hungarian texts, demonstrating its capability to handle language-specific nuances. Overall, this work contributes to natural language processing (NLP) and cyberbullying detection, showcasing a focused approach to applying machine learning in addressing societal challenges.