Methodology of a Swarm of Virtual Experts for Evaluating the Weight of Connections in Networks

Abstract

This article proposes a new methodology — the Swarm of Virtual Experts (SVE) — for evaluating the weights of connections in complex networks, based on a holistic approach. Traditional methods relying on expert assessments often face issues of subjectivity and limited resources. This paper introduces the methodology of the Swarm of Virtual Experts. The focus is on integrating large language models (LLMs) into the decision-making process, where each model acts as a virtual expert with specific tasks and functions. The core idea is to combine diverse assessments from different LLMs using mathematical tools, including incidence matrices, weighted averages, and aggregation methods. The methodology addresses the issue of fragmented results caused by the probabilistic nature of LLMs and enhances analytical efficiency through role assignment to agents, aggregation mechanisms, and quality evaluation of outcomes. The application of this technique is illustrated with examples, particularly in the field of cybersecurity. Special attention is given to holistic analysis, which provides a comprehensive approach to evaluating the weights of connections between nodes in networks.