Associative Intelligence for Object-Centric Process Mining with MPM (Extended Abstract) 1st Janna Meyer 2nd Josua Reimold 3rd Constantin Wehmschulte MEHRWERK GmbH MEHRWERK GmbH MEHRWERK GmbH Karlsruhe, Germany Karlsruhe, Germany Karlsruhe, Germany jm@mehrwerk.net jr@mehrwerk.net cw@mehrwerk.net Abstract—Complex processes nowadays are analyzed and im- a process instance. In a traditional event log, each event is proved with process mining software. Despite the high analytics related to one activity, one timestamp, and one case (i.e., a quality of commercial tools like MEHRWERK ProcessMining process instance). Taking a look at the example of a customer (MPM), process mining projects often face a substantial problem: the need to determine one case identifier to align all events into order given by van der Aalst [1], the concept of OCPM one process variant. This ”one case notion” complicates projects becomes clear: The process can be analyzed from different as it does not adhere to reality. Object-centric process mining perspectives - orders, items or packages. These objects are (OCPM) relaxes the ”one case notion” and allows the definition of interlinked, for example some items can be sent together in process instances by objects that interact with each other. Due to one package, others separately or even with another order. The this, OCPM represents processes more realistically. Commercial tools need to fully implement the concept of OCPM into their traditional process mining approach would only focus on one software functionalities to make its advantages of complexity perspective which can lead to certain disadvantages: reduction, simplification of data and event log modeling and thus gained understanding available to a vast group of industrial 1) Loops in the process graph that lead to huge complexity. users. Therefore, the software MPM is extended with OCPM Taking the order as case ID, the process step of ”picking and proposes first approaches to handle topics like conformance the item” will occur for every item in the order. checking, inter-object influence, or object-attribute analysis. 2) Duplicated events lead to skewed process performance Index Terms—Object-Centric Process Mining (OCPM), Con- indicators such as the number of process steps or au- formance Checking, MEHRWERK ProcessMining (MPM) tomation rates, if not carefully modeled. If the item is I. I NTRODUCTION picked as case ID, the process step ”place order” would occur for every item-case and thus suggesting a much The motivation to overcome the problems of traditional higher number of order creations. process mining originates from the work of Wil van der Aalst 3) The need to aggregate information. One order with [1], which is based on the concepts of multi-entity and artifact- multiple packages only has an aggregated delivery time centric process mining [2]. (traditional) instead of having multiple delivery cases MPM is (to our knowledge) the first self-service business with their respective time (OCPM). intelligence based process mining tool to include OCPM as a feature. A strong advantage is that it is based on the Qlik © The discovered object-centric directly-follows graph (DFG) Sense platform with its associative engine which provides an is shown in figure 1. The order to cash process displayed intuitive way of analyzing (multi-object) processes which is consists of three colored objects: orders - golden, items - shown in a collection of case studies1 . blue and packages - dark grey. Activities are colored like the The demo application created with the OCPM-data set object that causes them, arcs like the interacting object. When ”running example”2 is provided for testing purposes. The objects interact they make use of the same activity as can be mentioned data set was chosen to allow comparison with seen at the blue node ”pick item” which obviously originates other approaches, such as [1]. The OCPM feature was also from the item object but also is part of the package process, tested with SAP Order-to-Cash and Purchase-to-Pay, as well since a not-picked item cannot be put into a package. The as Oracle Invoice-to-Cash processes and several ready-to-use object-centric DFG is more comprehensive than the traditional event logs. The scalability for OCPM proved to be the same spaghetti-like DFG on the same event log. This underlines the as for traditional process mining. large advantage and is the main reason for developing OCPM for MPM. Through OCPM the number of process variants is II. OCPM FOR MPM - W HY AND H OW significantly reduced, since loops are largely eliminated and no Following the idea of OCPM, it is no longer necessary events from other object types can interfere. Thus, the need to determine one case ID for several objects that interact in to drill down into process data and analyzing only reduced 1 https://mpm-processmining.com/en/casestudy-access-en/ subsets is obsolete. This unlocks important information that 2 https://github.com/Javert899/pm4py-mdl/blob/master/example logs/mdl/ was previously hidden and could not be used to generate mdl-running-example.mdl actionable insights. Additionally, the frequencies shown on the Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). of objects bundled into a case. In the demo application this becomes clear as the overall process governance would have been 91.6% whilst the process governance for the order objects is 77.0%, 73.4% for the item and 90.5% for the package objects. Thus, it is shown that the traditional approach would not have detected the root cause of process deviations in item- and order-processing. As MPM is using its synergies with Qlik© , the other two objectives can be reached. Issues regarding data modelling such as the linkage of different objects can be easily tackled [4]. With the Qlik© Associative Engine it is possible to discover relations in even vast amounts of data due to a patented in-memory technology [5]. Following this, context information can be used unaggregated per object type. Until now developers had to decide whether to provide information per event (not necessarily available for entire case analysis and the filtering influenced the process graph) or as a fuzzy assignment to the traditional case ID. MPM-based OCPM gives the freedom to attach such information to an object case, whereas it can be available not only to the object itself but also to the other related objects. The demo video3 demonstrates how each object’s attributes are available for analysis and how the influence between the object types is assessed by only using the power of the associative engine for OCPM. Since certain analysis is dependent on the modelling of ob- ject interaction, MPM overcomes the downside of insufficient modelling by calculating the traditional model as well, e.g., to identify maverick buying (invoice posted before order) the invoice activity has to be defined in the order object, otherwise the pattern can’t be found easily. IV. C ONCLUSION Fig. 1. MPM’s object-centric process graph for three process objects The motivation for OCPM-based MPM is a more com- prehensive, less complex process analysis with simpler data and event modeling and thus, better quality. It simplifies pro- graph are correct. cess mining projects and increases the acceptance of process mining in organizations. To make OCPM enterprise-ready, III. I NNOVATION AND M AIN F EATURES traditional functionalities like conformance checking, the use Besides the fully interactive and customizable object-centric of object attributes for analysis and the investigation of inter- process graph - which is an adaption of existing research object relations have to be included which is presented in this results - the innovations intended to broaden the topic of demo. Further research and development is still required, like OCPM in the process mining community are the application simple information representation of the different objects to of conformance checking on object-centric event logs, the reduce the entry-barrier for OCPM-based process mining. provision and use of object attributes for analysis, and the R EFERENCES investigation of the influence between the object types. [1] W.M.P. van der Aalst, A. Berti, “Discovering Object-centric Petri Nets” The MPM Conformance Checking module [3] integrates Fundamenta Informaticae, vol. 175, no. 1-4, pp. 1–40, September 2020. seamlessly with OCPM, as it is able to consider different [2] D. Fahland, “Artifact-Centric Process Mining” Encyclopedia of Big Data case-properties when aligning the process model. Every object Technologies 2018, Springer, Cham. [3] J. Meyer, J. Reimold, Constantin Wehmschulte, “Conformance Checking type can be ”conformance checked” against its own or a Challenge 2019: Analysis of Central Venous Catheter Installation with global/traditional process model. The interaction of object MEHRWERK ProcessMining,” ICPM 2019 types can be taken into account by including the specific pro- [4] J. Meyer, J. Reimold, Constantin Wehmschulte, “An introduction to MPM - MEHRWERK ProcessMining,” CEUR Workshop Proceedings, cess steps in both objects. The influence of one object type’s vol. 2374, 2019 conformance to another one’s is related by the linkage of these [5] QlikTech International AB, “The associative difference: Freedom from objects. An object-centric conformance checking approach the limitations of query-based tools,” https://www.qlik.com/us/products/ associative-engine, 01.09.2021. derives more precise results because the process deviations can be allocated directly to single objects and not to a consortium 3 https://bit.ly/2YWqdbk Copyright © 2021 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).