We've built two systems that synthesise structural system dynamics models based on metadata (i.e., data described through a domain specific ontology). The first system, we call Synthesis-0, comprises synthesis constraints which define how to connect metadata to model structure. The system constructs models by, in an exploratory manner, finding metadata evidence to solve the constraints. In the second system, Synthesis-R, we take existing models (or reference models) as an additional synthesis resource. Instead of "blindly" exploring the metadata, Synthesis-R reuses a reference model to bound the search for metadata evidence. To this end, the constraints already defined in Synthesis-0 are harnessed for connections to be drawn between the reference model structure and the (new) metadata giving rise to new models. In other words, the reference model structure is matched (where possible) and marked up with new metadata.
We believed that the reuse feature in Synthesis-R should lead to shorter run times compared to Synthesis-0. To verify that we did an empirical evaluation of the systems' run times. In the talk I'll show the method applied in the experiment and the results obtained.