Application of large language models for intelligent expansion of semantic networks

This paper proposes a methodology for constructing and further expanding semantic networks based on text analysis using large language models (LLM). The initial semantic network is expanded with the help of GPT-4, Llama-3, and DeepSeek-V3, and the obtained results undergo quantitative evaluation of precision and recall. The proposed semantic networking technology is based on the concept of a "swarm of virtual experts", where each LLM enhances the semantic network with its own knowledge and connections. Additionally, an approach to network integration has been developed, enabling the consolidation of results from different LLMs into a unified structure. The use of a weak-link filtering mechanism enhances the reliability of the final network by eliminating false-positive connections and reducing information noise, contributing to a more accurate and complete representation of knowledge. For semantic network visualization, the Gephi software is utilized – a free and open-source network analysis and visualization tool.