Object-Centric Process Mining: Dealing with Divergence and Convergence in Event Data

Process mining techniques use event data to answer a variety of process-related questions. Process discovery, conformance checking, model enhancement, and operational support are used to improve performance and compliance. Process mining starts from recorded events that are characterized by a case identifier, an activity name, a timestamp, and optional attributes like resource or costs. In many applications, there are multiple candidate identifiers leading to different views on the same process. Moreover, one event may be related to different cases (convergence) and, for a given case, there may be multiple instances of the same activity within a case (divergence). To create a traditional process model, the event data need to be “flattened”. There are typically multiple choices possible, leading to different views that are disconnected. Therefore, one quickly loses the overview and event data need to be exacted multiple times (for the different views). Different approaches have been proposed to tackle the problem. This paper discusses the gap between real event data and the event logs required by traditional process mining techniques. The main purpose is to create awareness and to provide ways to characterize event data. A specific logging format is proposed where events can be related to objects of different types. Moreover, basic notations and a baseline discovery approach are presented to facilitate discussion and understanding.