S.517-522

Process mining allows one to analyze and extract knowledge from event data, i.e., records of process executions stored in information systems. Most process mining techniques are directly applied to the data as recorded in the system. Applying automated process discovery techniques, i.e., a core process mining technology, directly on such data yields complex process models describing millions of different execution paths. Other techniques applied to such discovered process models and system-level data, e.g., conformance checking or performance analysis techniques, often generate complex and over-detailed results. The results obtained by directly applying process mining techniques on system-level data are, therefore, hard to interpret by a human analyst and greatly differ from the business level. Therefore, in this paper, we propose a generic hierarchical framework for event abstraction. We formalize the framework, which uses the notion of activity instances as an input and allows for hierarchical abstraction of event data. In addition, we propose an instantiation of the framework, which describes two key functions of the framework, i.e., abstract concept identification and abstract entity extraction. The framework, together with the instantiation, is evaluated both quantitatively and qualitatively. The experiments show that, without compromising the quality of results, the abstraction allows users to easier analyze a process.