The article proposes a method for automated creation and modification of semantic networks using large language models (LLMs). The method uses formalized primitives to extract concepts, expand them, and combine them into a single network. For the first time, the concept of preferential semantic attachment is presented, which adapts the Barabashi-Albert model taking into account the semantic relevance of nodes. The experiments demonstrated the effectiveness of the proposed approach, using the example of the analysis of cybersecurity texts. The obtained results show the correspondence of the generated network to the power distribution, which is confirmed by the Kolmogorov-Smirnov criterion. The proposed approach can be used for information analysis and modeling of the evolution of knowledge in various subject areas. Keywords: semantic networks, large language models, prompt generation, preferred semantic association, information analysis, Barabashi-Albert model, cyber security, automated concept extraction, knowledge evolution