Аннотация
This research paper addresses the critical issue of cyberbullying detection within the realm of social networks, employing a comprehensive examination of various machine learning and deep learning techniques. The study investigates the performance of these methodologies through rigorous evaluation using standard metrics, including Accuracy, Precision, Recall, F-measure, and AUC-ROC. The findings highlight the notable efficacy of deep learning models, particularly the Bidirectional Long Short-Term Memory (BiLSTM) architecture, in consistently outperforming alternative methods across diverse classification tasks. Confusion matrices and graphical representations further elucidate model performance, emphasizing the BiLSTM-based model's remarkable capacity to discern and classify cyberbullying instances accurately. These results underscore the significance of advanced neural network structures in capturing the complexities of online hate speech and offensive content. This research contributes valuable insights toward fostering safer and more inclusive online communities by facilitating early identification and mitigation of cyberbullying. Future investigations may explore hybrid approaches, additional feature integration, or real-time detection systems to further refine and advance the state-of-the-art in addressing this critical societal concern.