Iraqi Dialect Cyberbullying Detection in Social Media Text Using Deep Learning Models

The prevalence of social media platforms has led to an upsurge in online interaction, and this has also allowed for harmful behaviors like cyberbullying. The problem is worsened in the case of Iraqi dialect due to the lack of linguistic resources with annotation and only limited Natural Language Processing (NLP) tools. The study of Arabic dialects is not as advanced as the study of the English language, and the automatic detection of abusive content is even more complex. The purpose of this study is to solve the aforementioned problem by creating a new dataset named Iraqi Dialect Social Media text from different sources which would be used to detect cyberbullying. A balanced set of 21,000 hand-annotated comments was created from Facebook, X and YouTube. Data preprocessing was done carefully and strictly following a well-defined data pipeline to ensure data quality and appropriateness for analysis. Furthermore, the performance of different deep learning models like Long Short-Term Memory (LSTM), stacked LSTM, Bidirectional LSTM (BiLSTM) and Hybrid CNN-LSTM model was tested. The experimental results demonstrate that the proposed models perform well, having an accuracy of more than 91.7% and the accuracy is achieved by the Hybrid CNN-LSTM model under all the evaluation parameters. The results show that the proposed model is effective, combining the advantages of the convolutional neural network and the long short-term memory network, to grasp the local linguistic properties and long-range sentence context of the informal dialectal text. A significant aspect of this work is the data preparation, where the data had a positive effect on the models performance. The results indicate that deep learning techniques can be useful for the effective detection of cyberbullying in low-resource dialectal setting, which can contribute to safer and inclusive online communication.