Construction of Fuzzy Cognitive Maps assisted by Artificial Intelligence (LLM): methodological process and financial use case

Fuzzy Cognitive Maps (FCM) are systems-modeling tools that represent weighted causal relationships between the concepts of a domain. Their traditional construction requires lengthy sessions with human experts, limiting their scalability. This article proposes and documents a methodological process to build FCM using Large Language Models (LLM) that act as domain experts. The process comprises three phases: (1) concept elicitation through structured prompts sent independently to multiple AIs; (2) interactive correction, weighting and grouping of concepts; and (3) generation, combination and export of the FCM. The process is illustrated with a real use case: the analysis of the influence of macroeconomic, market and risk factors on financial assets. The resulting maps are exported in visualization format (GraphViz .gv) and social-network-analysis format (Gephi .gdf), facilitating their use in multiple analysis environments. The advantages, limitations and future extensions of the process as a method to externalize and formalize mental models are discussed.