Ensemble Deep Learning (EDL) for Cyber-bullying on Social Media

Abstract

Cyber-bullying is a growing problem in the digital age, affecting millions of people worldwide. Deep learning algorithms have the potential to assist in identifying and combating Cyber-bullying by detecting and classifying harmful messages. This paper uses two Ensemble deep learning (EDL) models to detect Cyber-bullying on text data, images and videos—and an overview of Cyber-bullying and its harmful effects on individuals and society. The advantages of using deep learning algorithms in the fight against Cyber-bullying include their ability to process large amounts of data and learn and adapt to new patterns of Cyber-bullying behaviour. For text data, firstly, a pre-trained model BERT (Bidirectional Encoder Representations from Transformers) is used to train on cyber-bullying text data. The next step describes the data pre-processing and feature extraction techniques required to prepare data for deep learning algorithms. We also discuss the different types of deep learning algorithms that can be used for Cyber-bullying detection, including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep belief networks (DBNs). This paper combines the sentiment analysis model, such as Aspect-based Sentiment Analysis (ABSA), for classifying bullying messages. Deep Neural network (DNN) used the classification of Cyber-bullying images and videos. Experiments were conducted on three datasets such as Twitter (Kaggle), Images (Online), and Videos (Online). Datasets are collected from various online sources. The results demonstrate the effectiveness of EDL and DNN in detecting Cyber-bullying in terms of detecting bullying data from relevant datasets. The EDL and DNN obtained an accuracy of 0.987, precision of 0.976, F1-score of 0.975, and recall of 0.971 for the Twitter dataset. The performance of Ensemble CNN brought an accuracy of 0.887, precision of 0.88, F1-score of 0.88, and recall of 0.887 for the Image dataset. For the video dataset, the performance of Ensemble CNN is an accuracy of 0.807, precision of 0.81, F1-score of 0.82, and recall of 0.81. Future research should focus on developing more accurate and efficient deep learning algorithms for Cyber-bullying detection and investigating the ethical implications of using such algorithms in practice.