Automatically Combining Conceptual Models Using Semantic and Structural Information

A conceptual model is a necessary precursor to a simulation model. It defines the problem space by listing relevant concepts and it identifies salient mechanisms by specifying which concepts are related. In this paper, we focus on conceptual models expressed as causal or ‘cognitive’ maps, in which concepts form nodes and their relationships are directed, weighted edges specifying causal strengths. When modeling complex social systems, participants may provide their perspectives through individual causal maps. To create a simulation model, these maps need to be aggregated into a coherent conceptual model. Two challenges arise: nodes may have different names although participants ascribed the same meanings (i.e., linguistic variability) and disagreements on causal strengths need to be reconciled. Although other fields have long proposed algorithms to aggregate knowledge bases (e.g., ontology matching), there is a paucity of solutions for causal maps. In this paper, we propose a solution for causal maps that leverages recent advances in graph matching. We demonstrate the feasibility and potential of our approach on a case study with
$n=22$
maps.