Discovering Compact, Live and Identifier-Sound Object-Centric Process Models

The research area of object-centric process mining provides techniques to model and analyze business processes with interacting object types, such as orders, items and packages. An important step in the analysis of such processes is the automated discovery of a process model that represents the control flow and interaction of all object types. However, existing object-centric discovery algorithms often produce process models that are hard to interpret due to their exccessive complexity. Additionally, they often do not provide formal guarantees on important model properties, such as liveness and identifier-soundness. These properties guarantee, upon executing the model, that all participating objects can properly reach the end of their life-cycle and that all parts of the model can eventually become active. In this paper, we propose a new object-centric discovery algorithm to automatically construct compact object-centric process models that are guaranteed to be live and identifier-sound. For this purpose, we introduce object-centric process trees as an abstract view on object-centric Petri nets that provide both guarantees by construction. We evaluate our approach by applying it to a range of synthetic and real-life logs and find it to be feasible in terms of runtime and unique with regards to its provided guarantees.