Introduction

Process mining refers to the task of automatically discovering a model that describes a given set of log events registered for some process, and is reckoned as a valuable tool for supporting the (re)-design of workflows.

Different approaches have been proposed in the literature that mainly differ in the language used for representing process models, and in the specific algorithm used for discovering the best one - you can refer to the web site of the Process Mining community for an up-to-date state of the art in the field.

Following this line of research, we devised a strategy for producing a modular and accurate representation of the process, based on the discovery of different variants of the process itself.

The resulting process model is a hierachy of workflow schemata, that describes the process behavior at different level of refinement:

• at the highest level of refinement (leaves), each schema accurately model a specific class of executions (or process"variant" )
• any other schem offer a view over a set of process variants

In many cases, this allows to achieve higher levels of accuracy, especially if one aims at reducing the percentage of execution patterns that the mined model admits, but that are not registered in the log (soundness, or behavioral appropriateness).

An overview of the approach is presented in section Approach, while furhter details can be found in the paper Mining Expressive Process Models by Clustering Workflow Traces.