Unveiling Cyberbullying with a Character-CNN Model and Data Grouping for Enhanced Detection

Abstract

In today's digital age, the rise of social media has brought with it a troubling issue - cyberbullying. It has become a pressing social problem that calls for effective solutions, and one promising approach lies in the realm of machine learning. Our research focuses on tackling textual cyberbullying, as it is the most prevalent form of communication on social networks. However, dealing with social media content presents unique challenges. The texts are often short, noisy, and lack a structured format. Additionally, they may contain misspellings, symbols, and intentional distortions, making it difficult for traditional machine learning methods that heavily rely on vocabulary knowledge to perform well. To address these obstacles, we introduce a novel approach known as the Char-CNN (Character-level Convolutional Neural Network) model. Unlike traditional methods that work with words, our model operates at the character level. This means that the model learns from individual characters rather than entire words, making it more robust in handling spelling errors and intentional obfuscation present in real-world social media data. By leveraging the power of Char-CNN, we aim to accurately identify instances of cyberbullying in social media texts. This advancement holds significant promise in curbing the harmful effects of cyberbullying and creating a safer and more positive online environment for all users.