Distributed Discovery of Multi-Level Approximate Process Patterns

Process mining focuses on the discovery of knowledge about a (business) process from a set of its executions stored in an event log. Each event describes an activity and its performer. Process mining techniques allows automatically extracting the process model that gains insight into various perspectives, such as the control flow perspective, data, and organizational perspective. In this paper, we do not assume the existence of a single process model to which all process executions comply, and the goal is to discover a set of process patterns frequently occurring in the log. These patterns reveal the order of execution of events in a process, and, at the same time, they may describe properties of activities and performers participating to the events. Since activities and performers are often grouped in hierarchical categories, process patterns are discovered at multiple level of granularity. To deal with huge amount of data stored in a massive log, we resort to a distributed approximate discovery of multi-level frequent process patterns. Distinct process executions are distributed on several nodes of a Grid platform. Locally frequent patterns are discovered on each node. Global pattern set are then approximated and frequency is estimated on the basis of the local frequency of the pattern discovered at several nodes. Experiments are performed on a real event log.