Process Mining
Process mining automatically reconstructs models of business processes as they are actually lived, from the digital footprints in running IT systems. The quality of the discovered models is typically assessed along four dimensions: fitness, precision, generalisation, and simplicity. While fitness and precision are well-understood formally, simplicity, that is, how easily humans can understand a model, still lacks a universal definition; it depends on the use case and the users' prior experience.
My research focuses on empirically investigating this understandability specifically for workflow nets. Which structural properties, such as node count, density, connector degree, parallelism, separability, or properties of the reachability graph, correlate with it? In case studies with students at the University of Augsburg, I measure both objective (correctness of answers, time taken) and subjective understandability (pairwise comparisons between models).
To run these studies efficiently I built the tool "Petri-Dish", a platform that automates the creation, online delivery, and analysis of Petri-net surveys, feeding results directly into regression analyses. Early findings show, for example, that parallelism reduces objective comprehension, whereas the subjective rating is influenced more strongly by connector degree and separability.
