Abstract for: Can natural language processing methods identify subsystems of variables in system dynamics models?

This study seeks to answer the following research question: - Can natural language processing methods identify subsystems of variables in system dynamics models? - How consistent are NLP-based subsystem classifications with expert judgments? In this paper, variables from “A system dynamics model for national drug policy” were assigned to four subsystems availability, affordability, quality, and rationality by using Sentence-Transformers and cosine similarity. Variables with similarity differences below 0.1 were assigned to multiple subsystems, and the results were compared with the study. Subsystems are fundamental in system dynamics modeling, particularly in AI-assisted modeling, as they enable the incremental and structured communication of the modeling process to the machine. Based on our study, NLP methods can effectively assist in identifying subsystems however, further refinements and optimizations are required. Natural Language Processing can assist modelers in accurately assigning variables to their appropriate subsystems. Based on our study, out of 92 variables, 37 were assigned precisely to their correct subsystem, 8 were assigned to an incorrect subsystem. 26 variables were correctly assigned to at least one of their appropriate subsystems and 21 variables were assigned to at least one of their incorrect subsystem. translating- coding