Abstract for: From Text to Sector Maps: Automating System Dynamics Conceptualization Using Large Language Models

Sector maps (SMs) play a central role in system dynamics (SD) conceptualization by organizing variables into high-level structural domains and supporting the transition from qualitative problem descriptions to formal models. However, their construction remains largely informal, relying on expert judgment and lacking standardized conventions, which limits transparency, reproducibility, and comparability across studies. This paper addresses this gap by proposing a structured and automated approach for genera􀆟ng sector maps directly from textual descriptions of SD models. First, a rule-based structural framework for SM representation is introduced to reduce ambiguity in sector definitions and relationships. Second, a four-stage pipeline is developed that leverages large language models (LLMs) to translate unstructured text into sector-level representations by extracting variables, identifying causal relationships, grouping variables into sectors, and constructing sector maps. The approach is demonstrated using a published SD case study, where the generated sector structure is qualitatively and structurally compared with the original model. Results show that the method can recover meaningful sector groupings and interconnections from text, providing a reproducible alternative to manual sector identification. By formalizing the transformation from text to sector maps, this study advances methodological support for SD conceptualization and opens new opportunities for AI-assisted model development. translate unstructured text into sector-level representations by extracting variables, identifying causal relationships, grouping variables into sectors, and constructing sector maps