Abstract
In the context of current geopolitical instability and full-scale war, transport and logistics centers (TLCs) are becoming strategically important hubs whose resilience directly impacts national security. However, the growing number of hybrid attacks targeting the human factor through social engineering and insider activity creates new challenges for information security systems. Effective protection of TLCs at the current stage is significantly limited by the inadequacy of traditional monitoring tools against emerging threats implemented through text communications with a high level of semantic uncertainty. Keyword-only searching, without accounting for the structural logic of a dialogue, fails to identify complex manipulation scenarios where the threat is hidden within the context rather than specific terms. In view of this, the aim of the study is to develop a method for the intelligent analysis of TLC network content based on a combination of content analysis, conversation analysis, and vectorization algorithms using distance matrices to automate the detection of hidden threats. The synergy of these methods allows for both mass screening of messages and deep semantic verification of dialogue interaction. The mathematical framework of the method is based on the formalization of conversation "scenarios" through the construction of distance matrices Mj, where the elements represent relative distances between reference security concepts. To reduce the dimensionality of the resulting data, Principal Component Analysis (PCA) is employed, allowing for the formation of feature vectors for subsequent processing by a neuro-fuzzy network. Based on a linguistic rule base, the network calculates an integrated risk indicator R ϵ [0; 1], which serves as a trigger for activating preventive measures within security decision-support systems. Simulation results of legitimate and anomalous communication scenarios confirmed the high sensitivity of the method to structural deviations in dialogue logic, even when personnel utilize legitimate vocabulary. The analysis demonstrated a clear differentiation of risk levels for various types of threats, verifying the effectiveness of the developed mathematical framework. The practical significance of the results lies in the possibility of integrating the method into analytical protection modules for logistics systems, enabling security state assessment and proactive response to human-factor-related incidents. Prospects for further research involve studying the impact of AI agents on the dynamics of dialogue interaction.