Аннотация
The prevalence of abusive language and cyberbullying on social media platforms presents a growing challenge to the user safety and digital well-being, necessitating the development of effective automated content moderation systems. This study proposes a hybrid deep learning model that combines Long Short-Term Memory (LSTM) networks with Convolutional Neural Networks (CNNs) to classify the abusive text with enhanced accuracy and contextual awareness. The LSTM component captures long-range dependencies and semantic context, while the CNN module extracts discriminative local n-gram features. The model was trained and evaluated on three benchmark datasets: HatebaseTwitter, HatEval, and TRAC. The experimental results demonstrated that the proposed architecture outperforms traditional classifiers, such as SVM, Random Forest, and Logistic Regression, as well as standalone CNN and LSTM models, achieving superior performance across all standard evaluation metrics. Notably, the model attained AUC scores of up to 0.97, indicating robust discriminatory power. These findings underscore the effectiveness of the hybrid LSTM–CNN model for abusive language detection and highlight its potential for deployment in real-time content moderation tools aimed at fostering safer online communication environments.