Object Type Clustering using Markov Directly-Follow Multigraph in Object-Centric Process Mining

TL;DR

This paper presents a novel clustering method for case notions in object-centric process mining using Markov Directly-Follow Multigraphs, enabling more accurate and less assumption-dependent analysis of complex processes.

Contribution

It introduces a new clustering approach based on Markov Directly-Follow Multigraphs and a threshold tuning algorithm, integrated into a Python library for improved process model discovery.

Findings

Effective clustering of case notions demonstrated on P2P log

Clusters show high similarity in process behavior

Method reduces reliance on analyst assumptions

Abstract

Object-centric process mining is a new paradigm with more realistic assumptions about underlying data by considering several case notions, e.g., an order handling process can be analyzed based on order, item, package, and route case notions. Including many case notions can result in a very complex model. To cope with such complexity, this paper introduces a new approach to cluster similar case notions based on Markov Directly-Follow Multigraph, which is an extended version of the well-known Directly-Follow Graph supported by many industrial and academic process mining tools. This graph is used to calculate a similarity matrix for discovering clusters of similar case notions based on a threshold. A threshold tuning algorithm is also defined to identify sets of different clusters that can be discovered based on different levels of similarity. Thus, the cluster discovery will not rely on merely analysts' assumptions. The approach is implemented and released as a part of a python library, called processmining, and it is evaluated through a Purchase to Pay (P2P) object-centric event log file. Some discovered clusters are evaluated by discovering Directly Follow-Multigraph by flattening the log based on the clusters. The similarity between identified clusters is also evaluated by calculating the similarity between the behavior of the process models discovered for each case notion using inductive miner based on footprints conformance checking.