An Arabic Cyberbullying Detection System Using Convolutional Long Short-Term Memory Model (Dr. Maha Faisal)

Traditional methods for detecting cyberbullying often struggle with identifying instances in Arabic due to the intricacies of the language and its cultural context. In response, we developed a cyberbullying detection system specifically designed to analyze and classify Arabic comments on YouTube as either cyberbullying or not cyberbullying. Our system utilized a convolutional long short-term memory (ConvLSTM) model, tailored to effectively identify instances of cyberbullying in Arabic text. We compared our proposed ConvLSTM model against traditional machine learning models, such as Random Forest and Decision Tree classifiers, achieving superior performance metrics. The ConvLSTM model achieved an accuracy rate of 98.94% and an F1-score of 99%, significantly outperforming the traditional models. To provide a comprehensive analysis of different types of hate speech or cyberbullying, we created additional classifications for the dataset including General Hate Speech (GHS), Racism Hate Speech (RHS), Islamophobia, and Violence or Threats. Additionally, we developed a user-friendly interface (UI) to facilitate interaction with the detection system, ensuring that the detected results are accessible and actionable for users.

Supervisor: Dr. Maha Faisal
Convener: Prof. Ayed Salman
Examination Committee: Dr Ameer Mohammed