Development and Verification of an Information Security System for Protection Against Destructive Content Based on Transformer Models

Abstract

Development and Verification of an Information Security System for Protection Against Destructive Content Based on Transformer Models

Skvortsov A.A., Anureva M.S., Lopatin D.V., Koroleva N.L.

Incoming article date: 19.01.2026

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