The paper addresses the problem of automated detection of destructive verbal impacts in user-generated content of digital platforms as an element of information security assurance. A method for context-semantic identification of aggressive and discriminatory statements based on the RuBERT transformer model fine-tuned on a specialized annotated corpus of Russian-language messages is proposed. The procedures of data preparation, training of a binary classifier, and probabilistic interpretation of the results are described. Experimental evaluation confirms the effectiveness and robustness of the method with respect to lexical variability and context-dependent forms of verbal aggression, as well as the possibility of its integration into automated systems for monitoring and protection of the information space.

Keywords: information security, destructive content, verbal aggression, automatic moderation, context-semantic analysis, transformer model, RuBERT, binary classification, machine learning, natural language processing, monitoring system, intelligent filtering