AbstractMine: Privacy-aware Abstraction of Event Data for Distributed Process Mining

点击翻译按钮获取中文摘要

Process mining enables the analysis of process-oriented systems based on events recorded during their execution. However, existing techniques typically assume that an event denotes the execution of a single activity as part of a single case; and that all events are available as part of a log at a single location. Once process mining is adopted for events sourced from sensor networks, however, these assumptions are no longer valid. Events capture low-level information and first need to be abstracted to facilitate their interpretation in terms of the process to analyze. Also, events stem from distributed sources, which calls for explicit handling of data exchange between the involved entities. However, event abstraction and data exchange also open opportunities to address the emerging need for privacy-awareness in process mining. Controlling the level and location of event abstraction, guarantees on the privacy of involved process participants may be achieved, without imposing any assumption on the analysis to be conducted. In AbstractMine, we aim at providing the algorithmic foundations to incorporate privacy guarantees in the abstraction of events that are continuously generated by distributed sources. This includes a formalism to describe the distributed abstraction of events and link it to attacks on the privacy of process participants as well as the utility of the resulting data for operational analysis. Based thereon, we contribute algorithms for abstraction of events that give privacy guarantees, while maximizing the data utility. We consider these algorithms for centralized and distributed settings, as well as for static and streaming data. The collaborations within SOURCED enable us to link event abstractions to approaches for uncertainty management and explainability in process mining, as well as achieving scalability in particular in distributed application scenarios.