Event Abstraction for Partial Order Patterns.

Process mining endeavors to extract fact-based insights into processes based on event data stored in information systems. Due to the variety of processes in different fields and organizations, there does not exist a universal technique to allow for putting the process mining outcome directly into action. Various techniques have been developed to support human analysis. Meanwhile, as raw event data are often provided at the system level, the abstraction principle is applied to “lift” the data to a higher level for human interpretation, which is called event abstraction . Owing to the limitation of the information systems deployed in practice, most abstraction techniques are developed based on the assumption that all process activities are performed sequentially, ignoring the fact that there may be activities performed concurrently or the relation of the activity executions could not be clearly defined. In this paper, we propose an event abstraction framework based on partial order patterns. We extract the candidate pattern instances and abstract event data based on the pattern instances identified. Moreover, we instantiate the framework and optimize the implementation. The framework is evaluated with synthetic event data, and a case study based on a real-life process is performed, demonstrating the applicability of the framework.