Detecting Cyberbullying with Text Classification Using 1DCNN and Glove Embeddings

Abstract

In recent years, the number of social networking sites has exploded, providing a platform for individuals all over the world to connect and discuss their common interests. With the increased use of the Internet and social media platforms, it is no surprise that young people are utilizing these technologies to injure one another, and this happens on a variety of social media sites and apps. In a few of hours, a 10 s Snapchat post can cross platforms and go viral on Facebook, Instagram, and Twitter. The bulk of prior studies have used traditional machine learning models to tackle this challenge, and the bulk of the produced models in these researches are only adaptable to one social network at a time. Deep learning-based models have made their way into the identification of cyberbullying occurrences in recent studies, claiming to be able to overcome the limits of traditional models and increase detection performance. Given the negative effects of cyberbullying on victims, it is critical to identify effective ways to detect and prevent it. Deep learning can aid in the detection of bullies. The main goal is to identify the best algorithm for detecting cyberbullying using 1DCNN, Bidirectional Encoder Representations from Transformers (BERT), and Recurrent Neural Networks (RNNs) (RNN). In comparison with the other two models, 1DCNN and glove embeddings produce more exact results.