This paper addresses the critical issue of destructive content dissemination on social networks, particularly the disguising of propaganda as legitimate material and the heroification of extremist ideologies. It analyzes existing content analysis methods and proposes a novel approach that integrates graph representation of information objects in social networks with color coding for visual evaluation. This integrated system leverages graph representation to display relationships between information objects and color coding to simplify the evaluation of content tone and context. A key strength of the proposed system lies in its versatility and adaptability for contextual analysis across diverse practical applications. The paper presents the development of a system based on BertForSequenceClassification model and demonstrates its application in analyzing posts from VKontakte, showcasing the potential for improved detection and prevention of harmful content. The study concludes by outlining future research directions, including enhancements to semantic analysis, incorporation of temporal dynamics, and the exploration of advanced clustering and graph centrality metrics to improve the accuracy and reliability of the method.