Unmasking Cyberbullying in Bengali: A Deep Dive into Digital Harassment

Cyberbullying is a growing problem online, and it's not just in English. In the Bengali-speaking world, it's a big issue too. But there's not much research on it. That's why a recent study looked at over 70, 000 social media comments in Bengali. The goal? To understand and detect cyberbullying better. First, they cleaned up the data. They sorted comments into positive and negative ones. Then, they used a method called Latent Dirichlet Allocation (LDA) to find patterns in the negative comments. They looked at things like age, gender, ethnicity, religion, and other topics.Next, they tested different models to see which one could detect cyberbullying the best. They tried Support Vector Machine, XGBoost, and even some fancy ones like CNN+BiLSTM+GRU and mBERT. mBERT did the best, with 92% accuracy. But the hybrid model CNN+BiLSTM+GRU wasn't far behind, at 91%. They didn't stop there. They tweaked the models by adding BERT embeddings to CNN and ANN. That boosted the accuracy to 93%. To make sure the models were fair and clear, they used a method called Local Interpretable Model-agnostic Explanations (LIME). This helped explain how the models made their predictions. This study is a big step forward. It shows that cyberbullying in Bengali can be detected accurately. But there's still more to do. Cyberbullying is a complex issue, and it's important to keep improving how we tackle it.