FAEO-ECNN: cyberbullying detection in social media platforms using topic modelling and deep learning

The widespread use of Social Media Platforms (SMP) such as Twitter, Instagram, Facebook, etc. by individuals has recently led to a remarkable increase in Cyberbullying (CB). It is a challenging task to prevent CB in such platforms since bullies use sarcasm or passive-aggressiveness strategies. This article proposes a new CB detection model named FAEO-ECNN for detecting and classifying cyberbullying on social media platforms. The proposed approach integrates Fuzzy Adaptive Equilibrium Optimization (FAEO) clustering-based topic modelling and Extended Convolutional Neural Network (ECNN) to enhance the accuracy of CB detection process. Initially, pre-processing is performed in order to cleanse the dataset. Next, the features are extracted using multiple models. The unsupervised Fuzzy Adaptive Equilibrium Optimization (FAEO) is utilized for discovering the latent topics from the pre-processed input data, which automatically examines the text data and creates clusters of words. Finally, the cyberbullying classification makes use of the ECNN and Rain Optimization (RO) algorithm to detect CB from posts/texts. We evaluated the proposed FAEO-ECNN thoroughly with two short text datasets: Real-world CB Twitter (RW-CB-Twitter) and Cyberbullying Menedely (CB-MNDLY) datasets in comparison to State of The Art (SoTA) models like Long Short Term Memory (LSTM), Bi-directional LSTM (BLSTM), RNN, and CNN-LSTM. The proposed FAEO-ECNN model outperformed the SoTA models in detecting Cyberbullying on SMP. It has obtained 92.91% of accuracy, 92.28% of recall, 92.53% of precision, and 92.40% of F-Measure over CB-MNDLY dataset. Moreover, it has achieved 91.89% of accuracy, 91.32% of recall, 91.81% of precision, and 91.56% of F-Measure on RW-CB-Twitter dataset.