Configurable event correlation for process discovery from object-centric event data

Many modern enterprises are employing serviceoriented systems to execute their transactions. These systems generate an abundance of events, which can be analyzed to diagnose and improve business processes. However, the events are distributed over different data sources, as service-oriented systems are often implemented asWeb services involving different departments in an organization. They need to be consolidated in a process view since it makes no sense to analyze individual events. Existing approaches depend on an explicit case notion to correlate events. They work well on data from process-aware systems (e.g., BPM/WFM systems) which record events explicitly and each event is attached with a case id (a global identifier) to indicate its related process instance. However, most serviceoriented systems (e.g., ERP and CRM) produce object-centric data (e.g., database tables), which record events implicitly (e.g., through redo logs) and separately without a common case id. In this paper, we propose a novel approach to correlate events by the data perspective (e.g., by a so-called 'object path'). Besides, we define the concepts of correlation patterns and evaluation metrics, which enable users to evaluate and select the best way to correlate events based on their needs.