Object-Centric Process Mining using Celonis
Student: Louis Herrmann (BSc)
Title: Object-Centric Process Mining using Celonis
Supervisor: Alessandro Berti
1st Examiner: Prof. Wil M.P. van der Aalst
2nd Examiner: Prof. Ulrik Schroeder
Classical process mining approaches enforce the requirement of having only one case notion per event. However, in reality, event data often is in an object-centric format where this restriction is violated. Forcing object-centric data into this restrictive flat format can lead to a multitude of problems such as the duplication of events and inaccurate frequency information. To bridge this gap between objectcentric and case-centric process mining, we establish a link between the case-centric data stored in a Celonis Data Model and an object-centric event log (OCEL). We accomplish this by introducing a set of operators that allows us to transform an OCEL to a Celonis Data Model and vice versa. During the flattening process, the exact object-object and event-object relationships can be lost. To combat this loss of information, we let the user induce additional information that they might have about the process into the transformation back to an OCEL. The advantages of transforming flat data into an object-centric format is that it allows for the (re- )discovery of the true object-object and event-object relationships which can help us better understand the process at hand. Furthermore, we propose a new technique to abstract event data on an objectcentric level. This has the benefit that we can incorporate information into the abstraction process that is not available in a case-centric setting and thus end up with more accurate high-level data models. We pack the proposed operators into an intuitive and easy to use UI so that no technical knowledge of a query language such as SQL or PQL is required to efficiently perform the transformation. To test the proposed methods, we perform experimental analysis on a publicly available OCEL and demonstrate that with this novel approach we can attain results that were not previously possible with traditional flat-based techniques.