Process Mining using Parameterised Graphs

This is a brief report on Andrey Mokhov’s visit to Josep Carmona’s research lab in Universitat Politècnica de Catalunya. The report summarises potential benefits that Parameterised Graphs (PGs) [1] can bring to the domain of process mining [2] and outlines important challenges to be addressed in future research. The authors intend to elaborate on the presented ideas in followup peer reviewed publications. Andrey Mokhov, Josep Carmona: Process Mining using Parameterised Graphs 1 Motivation for using PGs in process mining In this section we discuss key reasons that motivate us to study the application of PGs in process mining, namely: (i) their ability to compactly represent complex behaviours without excessive over-generalisation, and (ii) the possibility of capturing event log meta data as part of PG representations and taking advantage of the data for the purpose of explaining the processes under observation. Note that in this report we work with PGs whose equivalence is considered up to the transitive closure. Such PGs have historically been studied under the name of Conditional Partial Order Graphs [3]. 1.1 Compact representation We have performed a number of experiments with state-of-the-art process mining software and found out that the existing solutions often produce too general solutions for event logs corresponding to processes with nontrivial mix of concurrency, causilty and choice. One example of such an event log is L = {abcd,cdab,badc,dcba}. One can notice that in this example the order between events a and b always coincides with the order between events c and d, which we consider an important piece of information about the process. The existing process mining methods, however, fail to capture this information and produce the most general explanation for the process in which all four events are concurrent. x