Technology Blogs by SAP
Learn how to extend and personalize SAP applications. Follow the SAP technology blog for insights into SAP BTP, ABAP, SAP Analytics Cloud, SAP HANA, and more.
cancel
Showing results for 
Search instead for 
Did you mean: 
robert_bluemel
Advisor
Advisor
628
In the heart of Rome, the fifth edition of the International Conference on Process Mining (ICPM) took place from October 23 to 27. Despite being relatively young, the conference has gained its reputation as one of the core venues for process mining alongside the BPM and the CAISE conferences. Dedicated exclusively to process mining, the conference unfolds over three focus areas – innovative techniques, fundamental research, and practical applications and case studies – aiming to bridge between academia and industry.


This year’s edition once again attracted more than 400 people worldwidefrom all over the world to attend the 2023 ICPM. The above figure depicts the distribution of participants. It shows that about one half comesis coming from academia, whereas the remaining half comprises representatives from the industrial sector. The latter can in fact further be subdivided (roughly) into thirds representing vendors (such as e.g., SAP Signavio, Celonis, Microsoft, and Apromore), consultancies (mainly represented by EY), and practitioners (from the software industry and other sectors).

These various interests that are combined in the ICPM were also reflected by SAP Signavio’s presence at the conference . On the one hand, colleagues from various teams and backgrounds, such as research, development, or senior management, joined to learn and to participate. On the other hand, being a sponsor of the conference, SAP Signavio maintained a stall at the conference location that remained a crowded spot throughout the breaks of the conference, attracting both: researchers (affiliated or interested) and customers.


Between these breaks the conference offered a wide range of program combining research and industry in various sorts of formats: spanning from research presentations at a set of workshops and the main research track, over demo presentations, to an entire industry day, mainly consisting of discussions.
The topics that were discussed during these sessions met the general conference’s aspiration for providing and discussing research not only on current core process mining topics such as process discovery and conformance checking. Also, more fundamental topics, such as stochastically known logs and object centric process mining, were discussed as well as industry collaborations. While machine learning has been a prominent trend in the conference overreflected in the conference throughout the past few years, statistical methods gained more attention in this year’s edition.
SupposeIf you are interested ininto more details about the concepts and innovations that were presented, as well as how the spaghetti process metaphor was pushed beyond the boundaries of good (Italian) taste. In that case,, then you are welcome to continue reading.

Research at ICPM


The research track was opened by a fascinating keynote from Avigdor Gal titledwith the title “Everything there is to know about stochastically known logs”. The talk emphasized the need for a paradigm shift in seeingthe way we see process logs. While back in the days data was stored in well-organized, centrally managed, and carefully monitored organizational information systems, we have seen a tremendous change to massive amounts of data being collected from sensors or generated by AI. Thus, data qualitythe quality of data cannot be guaranteed, which also holds for process data. Therefore, the author argues for seeing event logs no longer in a deterministic manner, but inspecting the log through the lens of uncertainty, motivating the need for stochastically known logs through modern process mining applications. Concretely, this means that, for example, each activity in the trace is assigned a certain probability of being executed, if at all.

Following the keynote, variousa wide variety of papers were presented, ranging from process discovery, over simulation to AI in process mining. While one presentation stood out because it depicted pineapple on top of a plate of spaghetti (processes), the paper that generally resonated best and won the best paper award was titled “Addressing the Log Representativeness Problem using Species Discovery.” The work of Kabierski et al., which also has a statistical approach, is concerned with the extent to whichquestion to which extent the log captures the characteristics of the process that are relevant for the analysis. The authors tackle this question using estimators from biodiversity research. Using real-life event logs, the results indicate that most often the logs cannot be assumed to entirely represent the underlying process, indicating that they lack critical information needed for trustworthy conclusions. This finding stresses the need for future research to represent the underlying process more faithfullymore faithfully represent the actual underlying process.
For more details on the presented papers, we refer the reader to the proceedings of ICPM 2023.

One major topic that was addressed in several publications and which we want to emphasize here is Object-Centric Process Mining (OCPM). The importance of it to the community was also reflected in the symposium that was held to discuss developments in Object-Centric Event Data (OCED) at the conference, since a first OCED data model was presented last year. What stood out was that one contribution was presented as the new Object-Centric Log Standard. The symposium gave insights into how a simplified version of an OCED-like meta-model can be used in conjunction with OCPM to disentangle processes, figuratively displayed by pictures of differently colored spaghettis. However, the following discussion showed that there was no consensus on the industry-scale usefulness of OCED as a standard on par with, for example, BPMN; similarly, practical perspectives on OCPM, particularly concerningwith respect to economic scalability and business value of generated insights, remain open questions.

The official research track of the conference was accompanied by six different workshops: Machine Learning for Process Mining (ML4PM), Process Querying, Manipulation, and Intelligence (PQMI), Process-Oriented Data Science for Healthcare (PODS4H), Event Data and Behavioral Analytics (EDBA), Education meets Process Mining (EduPM), and Collaboration Mining for Distributed Systems (COMINDS).
The ML4PM workshop was opened by a keynote from Jochen de Weerdt, pointing out some critical challenges in predictive process monitoring that might lay the foundation for future research. One of the aspects was concerned with the ongoing gap between academia and industry. The identified pathways to increase real-life adoption include incorporating uncertainty and model interpretability, increasing model robustness, and moving from predictive to prescriptive process monitoring.

The Industry Day


Process mining thrives as an area of applied research, so the ties to industry are important. This year, the industry day spanned two days, making space for a “post-industry” party for networking.
During the day, the industry focus involved changes in the organization of the exchange formats, brought different people on the stages, and different topics on the agenda. So mostly vendors, consultants and practitioners were in charge of discussions and topics such as AI were represented more widely.

The beginning of the industry day was delivered by Marlon Dumas, Professor and co-founder of Apromore, with his keynote on “Walking the Way from Process Mining to AI-Driven Process Optimization.” Based on a manifesto on AI-augmented business process management systems he mapped the well-known concepts of analysis steps, namely descriptive, predictive, and prescriptive analysis, to the process mining domain. With that, he outlined that current predominant analysis methods would only suffice for a descriptive analysis and that more predictive methods, such as predictive monitoring and what-if analyses, would be required to reach further optimization. He went on that only with the help of AI methods domain knowledge could be incorporated as context to even come to a prescriptive level.

What followed were three different types of discussions: smaller interactive group discussions, a so-called Masterclass of selected members, and a panel discussion with representatives from the three groups: vendors, consultancies, and practitioners. The first set of discussions was designed to cover specific topics, such as cultural, collaborational or technical aspects of process mining. In the latter session, co-hosted by SAP Signavio’s Senior Director of Product Management for Process Intelligence Silvio Arcangeli, the discussion centered on storytelling that could increase the acceptance of the data collection steps and how AI could be used as a sparring partner to conduct projects.


In contrast to that, the following discussion occurredtook place only among a selected group of participants of which Ron Agam, SAP Signavio’s CPO, was one of them. The focus of this Masterclass was to discuss how the general process mining market could be increased. Part of that was how trends could shape this development. Certain and certain aspects of process mining, such as the multi process perspective or data anonymization, could be demystified to increase user acceptanceacceptance among users.
The panel discussion that then closed the industry day again reiterated on previously discussed topics such as storytelling, AI, and OCPM. For the first topic, all six participants agreed that not tools or end results, but journeys are to be sold. For the remaining topics, soon most participants could agree on AI (for the moment) being a tool to make insights more available for everybody and that some object-oriented process mining could be beneficial in certain settings.

Concluding the Conference


To conclude, once more the ICPM succeeded in being a vivid place for exchanging new research in process mining and discussing future directions in both academia and industry.
That these directions are not completely aligned show the two examples of OCPM and GenAI:
While discussed only partially during the industry day, OCPM was widely present during the research track, mainly pushed by a specialized group of academics . In contrast to that, generative AI was not a focus in the research track, but heavily discussed during the industry day.
Of course, these directions do not have to be aligned but observing where they depart and where they overlap remains exciting, and where else to study this than at the ICPM?

 

Co-authored by Lukas Kirchdorfer and Robert Blümel