Deep Learning for Multi-Labeled Cyberbully Detection: Enhancing Online Safety

Social media platforms offer undeniable benefits, but the preservation of anonymity has led to the emergence of cyberbullying, a concerning social problem. This form of online harassment creates a negative and hostile environment, resulting in decreased user engagement and psychological harm to victims. According to ResearchGate and ScienceDaily, cyberbullying victims in the United States are 1.9 times more likely to commit suicide, highlighting the severity of the issue. However, the current research on cyberbullying detection has been limited to binary/multi-class text classification due to the lack of comprehensive datasets for training and evaluation. To address this gap, we developed a DL-based multi-labeled cyberbully detection system using a dataset of 95,608 social media comments. These comments were categorized into five distinct multi-labeled classes, allowing for a more comprehensive understanding of the different dimensions of cyberbullying. We utilized DL architectures, such as LSTM, BiLSTM, CLSTM, and BiGRU, to develop advanced cyberbully detection systems. By comparing the performance of these DL models with the ML models, we were able to assess the effectiveness and superiority of DL approaches in accurately identifying instances of cyberbullying contents. The CLSTM model, outperformed the others with an exceptional binary accuracy of 87.8% and a macro f1-score of 88.3%. CLSTM's ability to integrate both local and sequential information, coupled with its capacity to capture complex patterns and long-term dependencies, contributes to its superior performance in identifying and classifying cyberbullying instances. By successfully identifying and preventing cyberbullying, our study can contribute to creating a safer and more positive online environment, ultimately enhancing user engagement and satisfaction.