A Methodology for Aligning Process Model Abstraction Levels and

Stakeholder Needs

Dennis G. J. C. Maneschijn, Rob H. Bemthuis, Faiza A. Bukhsh, and Maria-Eugenia Iacob

University of Twente, Drienerlolaan 5, 7522 NB, Enschede, The Netherlands

Keywords:

Process Mining, Abstraction Levels, Stakeholder Analysis, Process Models.

Abstract:

Process mining derives knowledge of the execution of processes through analyzing behavior as observed from

real-life events. While beneﬁts of process mining are widely acknowledged, ﬁnding an adequate level of

detail at which a mined process model is suitable for a speciﬁc stakeholder is still an ongoing challenge.

Process models can be mined at different levels of abstraction, often resulting in either highly complex or

highly abstract process models. This may have an important impact on the comprehensibility of the process

model, which can also differ from the perspective of a particular stakeholder. To address this problem from

a stakeholder-centric perspective, we propose a methodology for determining an appropriate level of process

model abstraction. To this end, we use quantitative metrics on process models as well as a qualitative eval-

uation by using a technology acceptance model (TAM). A logistics case study involving the fuzzy process

mining discovery algorithm shows initial evidence that the use of appropriate abstraction levels is key when

considering the needs of various stakeholders.

1 INTRODUCTION

Process mining helps to unveil actionable knowl-

edge and insights of a process, based on historical

execution data in the form of event logs (Van der

Aalst, 2016). Process mining aims to discover (i.e.,

learning a model from behavior recorded in an event

log), monitor, and improve processes based on event

logs. Process mining, as a relatively young disci-

pline, could provide valuable insights in, for exam-

ple, performance evaluation, root-cause analysis, bot-

tleneck analysis, and process prediction and optimiza-

tion (Van der Aalst, 2016).

Although process mining has the potential to pro-

vide many promising insights and to support pro-

cess improvements, there are still many challenges

to overcome (Dos Santos Garcia et al., 2019). For

example, research on bottleneck analysis techniques

utilizing process mining is still an under-researched

area (Bemthuis et al., 2021b). Another challenge

concerns ﬁnding an appropriate balance between a

process model that is comprehensible, but still suf-

ﬁciently detailed in order to show relevant behavior

to the user of the model (Van der Aalst and Gunther,

2007; Leemans et al., 2020). Sometimes, the gen-

erated process models tend to be confusing and dif-

ﬁcult to understand, especially when there are many

diverse cases with deviating behaviors. Although this

complexity can be useful for some stakeholders inside

an organization, generally speaking, applying no ab-

straction could make the model too complex to com-

prehend (Van der Aalst, 2016).

To deal with process model complexity, abstrac-

tion can be applied, which can make the process

model less spaghetti-like and more comprehensible.

Abstraction omits lower-level information, which is

insigniﬁcant in the chosen context, from the visual-

ization (Van der Aalst and Gunther, 2007). For exam-

ple, typically not all small roads and pedestrian paths

are shown to bus drivers when considering a city road

map since it would make the map cluttered and incon-

venient to use. We consider the following deﬁnition

of abstraction: “Simplifying process models by re-

moving edges, clustering nodes, and removing nodes

to make the process model more suitable for the per-

son looking at it.” In other words, abstraction is about

the level of granularity of the process model.

A model can have a too high abstraction (under-

ﬁtted) or a too low abstraction (overﬁtted) (Van der

Aalst and Gunther, 2007). Finding an appro-

priate level of abstraction also depends on which

user/stakeholder is using the model, because users

typically have different needs and purposes. Yet, pro-

viding a process model that is suitable for a des-

Maneschijn, D., Bemthuis, R., Bukhsh, F. and Iacob, M.

A Methodology for Aligning Process Model Abstraction Levels and Stakeholder Needs.

DOI: 10.5220/0011029600003179

In Proceedings of the 24th International Conference on Enterprise Information Systems (ICEIS 2022) - Volume 1, pages 137-147

ISBN: 978-989-758-569-2; ISSN: 2184-4992

137

ignated stakeholder can be challenging. For ex-

ample, in the logistics domain where many stake-

holders have different needs, experiences, and back-

grounds (Flod

en and Woxenius, 2021; Tolentino-

Zondervan et al., 2021). Real-world processes of-

ten involve unstructured and ad-hoc behavior, which

produce spaghetti-like models. It seems reason-

able to argue that simpliﬁcation and abstraction are

needed at higher levels of management, but especially

those working ‘in the trenches’ of a process not only

must know about the details, they can often also tell

you why those details are there and whether or not

these steps are needed or can be circumvented/re-

engineered.

Multiple papers attempt to address the issue of

abstraction in process mining (G

unther and Van

Der Aalst, 2007; Baier et al., 2014; Kumar et al.,

2017; Van Cruchten and Weigand, 2018b). How-

ever, to our knowledge, none of these works try to

seek an abstraction level that serves the purpose of

a particular stakeholder. Instead of applying abstrac-

tion on process models directly (e.g., through ﬁlter-

ing), abstraction has also been applied to the level

of event data. A literature review on event abstrac-

tion already showed the importance of applying pre-

processing techniques for the successful application

of process mining (Van Zelst et al., 2021), such as

in large-scale industrial ERP systems. Nevertheless,

as mentioned by these authors, as well as concluded

by research on classifying event abstraction articles

(Diba et al., 2020), today’s approaches still often rely

on strong assumptions and domain experts.

We acknowledge the importance of applying ab-

straction at the level of pre-processing, but signif-

icant beneﬁts can also be obtained once a process

model has been created, because sometimes it might

not even be possible to apply pre-processing tech-

niques on event records (for example, when dealing

with noisy or incomplete data sets) (Zakarija et al.,

2015; Van Zelst et al., 2021). Alternatively, it could

be the case that we only have the process model dis-

tilled from many events and that we do not, or only

to a limited extent, have information about the exact

event records that were used to generate the process

model. For example, due to privacy concerns, only

some of the original event log’s content may be re-

vealed (Fazzinga et al., 2018b). Hence, it may not be

possible anymore to gather the original event logs.

In the realm of process model complexity, paral-

lels may be drawn with the discipline of enterprise

architecture (EA). Many stakeholders involved in EA

have a different perception of the complexity of an

architectural model (Iacob et al., 2018). Further-

more, EA models are often showcased from a partic-

ular viewpoint which could be speciﬁcally designed

for a target group. For example, business executives

may focus on its value delivery, management on its

functionalities and costs, IT architects on its maintain-

ability, and software developers on its ﬂexibility (Ia-

cob et al., 2018). In turn, these different perceptions

may lead to disagreement and mismanagement. Sim-

ilar reasoning has been mentioned regarding business

process model complexity metrics (Gruhn and Laue,

2007; Muketha et al., 2010). The notion of using both

objective and subjective metrics for striving towards

an optimal level of complexity to effectively and efﬁ-

ciently understand and use process mining, is an area

we aim to further explore.

To summarize, the beneﬁts of using process min-

ing are widely acknowledged, but the generated pro-

cess models can be challenging to understand for

(business) users (Yazdi et al., 2021). Therefore, we

may not achieve the (intended) goals of process min-

ing because we are unable to review the as-is pro-

cess models in comparison to the to-be process model

(to a sufﬁcient degree). To address this shortcom-

ing, this paper proposes a methodology that deals with

model complexity, and aligns it with the needs of var-

ious stakeholders. We focus on the application of ab-

straction within the scope of the discovery phase of

process mining, as discussed in the process mining

project methodology of (Van Eck et al., 2015). Since

our main research product (i.e., design artifact) is a

methodology, we followed Peffer’s Design Science

Research Methodology (DSRM) (Peffers et al., 2007)

during this research study.

The remainder of this paper is structured as fol-

lows. Section 2 presents the proposed methodol-

ogy followed by Section 3 where the methodology is

demonstrated by using a logistics case study. Sec-

tion 4 discusses the related work. Finally, Section 5

concludes and gives some pointers to future work.

2 A METHODOLOGY FOR

ALIGNING PROCESS MODEL

COMPLEXITY AND

STAKEHOLDER NEEDS

The proposed methodology in this paper deals with

the problem of making process models comprehensi-

ble for organizational stakeholders. The methodology

consists of the six phases as shown in Figure 1, each

requiring a certain input and delivering an output. Our

methodology is inspired by CRISP-DM (Wirth and

Hipp, 2000), a well-known methodology for data sci-

ence projects. In the remainder of this section, we

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

138

brieﬂy explain each phase.

Phase 1 - Business Understanding & Data Prepa-

ration. The ﬁrst phase forms the basis for conducting

the other phases of our methodology. It allows the

user to get familiar with the domain in which the re-

search is conducted. Furthermore, it requires the user

to get to know the data to be worked with and which

preparation steps were conducted.

Phase 2 - Identifying Relevant Experts/

Stakeholders. The second phase uses the busi-

ness understanding to deﬁne the relevant ex-

perts/stakeholders for which one wants to determine

the appropriateness concerning the generated pro-

cess models. Apart from deﬁning the relevant

experts/stakeholders, it is also important to identify

the (business) goals of each expert/stakeholder.

Phase 3 - Establishing Abstraction Levels. Using

the stakeholders and business purposes from phase

2, phase 3 deﬁnes the different abstraction levels at

which we aim to produce the process models. This

can be done by considering the characteristics of the

chosen process miner, in combination with the data

understanding and expert/stakeholder overview. The

goal is to have a distinct separation of organizational

levels represented through speciﬁed abstraction lev-

els.

Phase 4 - Deﬁning Quality Measurements. Before

generating the process models, it is important to de-

termine the quality measures that will be used in the

evaluation phase. It is recommended to determine

both quantitative and qualitative measures. In the end,

our methodology produces a social impact since it at-

tempts to ﬁnd a solution to a model complexity prob-

lem faced by stakeholders. Therefore, we highlight

the importance of using a quantitative and qualitative

measuring tool.

Phase 5 - Modeling. In this phase, we select and ap-

ply an algorithm for process model extraction.

Phase 6 - Evaluation. The ﬁnal phase uses the stake-

holder overview, quality measurements, and process

models to evaluate the quality of the process mod-

els for each stakeholder using an adaptation of the

Technology Acceptance Model (TAM) (Davis, 1989).

In the end, the evaluation provides an alignment of

stakeholder needs with process model complexity.

3 DEMONSTRATION

The case study used to demonstrate our methodology

is introduced in earlier work (Bemthuis et al., 2020).

This study concerns the transport of perishable goods

in a production facility. Multiple data sources, such

as smart pallets, are used to collect data about the

state of the shipments, transport units, and the en-

vironment (Bemthuis et al., 2020). The production

facility strives towards a good balance between min-

imizing quality decay against minimizing operating

costs. Investigating this case study seems promising

as multiple stakeholder perspectives could be consid-

ered as well as complexity levels, because of the pub-

lic data set (Bemthuis et al., 2021a) and because many

of the partners involved in the previous work are also

involved in the present research. This section will ap-

ply each phase of our methodology to this case study.

3.1 Business Understanding & Data

Preparation

The business understanding mainly included under-

standing the case study and getting to know character-

istics of the logistics domain, as well as the challenges

this brings forward in process mining. Logistics pro-

cesses are known for being complex and dynamic, of-

ten producing spaghetti process models, which seems

an interesting domain for our study.

The data preparation phase consists of enriching

the event logs of the case study. The event logs de-

scribe activities about movable transport units that

transport smart pallets from one place to the other.

The pallets and transporters are equipped with sen-

sors that keep track of the status of products and/or

transporters (Bemthuis et al., 2020). The quality of

products decays over time. How fast a product quality

is depreciating, depends on, e.g., the type of vehicle

and type of food transported. The data set comprised

several scenarios and experimental runs per scenario

(Bemthuis et al., 2021a). We considered the scenario

and experimental results of which the average product

decay was the lowest and of which a warm-up period

was removed. More details about the case study can

be found in (Bemthuis et al., 2020).

Two additional attributes of the event log were

used: the decay level (DL) of a product and a vehicle

identiﬁer. The DL is recorded as a numerical value.

To make the stakeholder assessment comprehensible

for the stakeholders, we classiﬁed the DL into four

categories. The categories can be found in Table 1.

These categories are based on the mean and the stan-

dard deviation of the DL per scenario. We have de-

cided to not use predeﬁned quality levels, because we

aimed to obtain events that are fairly distributed over

the different quality levels (i.e., proportionally for the

scenario). After enriching the data with this categori-

cal value, we ﬁltered the event log to remove incom-

plete traces. Filtering was done using the heuristics

ﬁlter plug-in of the open-source tool ProM.

A Methodology for Aligning Process Model Abstraction Levels and Stakeholder Needs

139

Raw data

Processed

data

Stakeholder overview

Process models

Abstraction

levels

Evaluation method

Alignment of stakeholder needs

and model complexity

Phase 1.

Business understanding

& data preparation

Phase 2.

Identifying relevant

stakeholders

Phase 3.

Establishing abstraction

levels

Phase 4.

Defining quality

measurements

Phase 5.

Modeling

Phase 6.

Evaluation

Figure 1: Methodology followed for aligning stakeholders to process model abstraction levels.

Table 1: Product quality decay categories used for data en-

richment.

Quality category Partition of

Good DL ≥ µ + σ

Sufﬁcient µ ≤ DL < µ + σ

Insufﬁcient µ − σ < DL < µ

Poor DL ≤ µ − σ

3.2 Identifying Relevant Stakeholders

Actively managing stakeholders and addressing the

needs of stakeholders is beneﬁcial for an organization

(Greenley and Foxall, 1997; Post et al., 2002). As

mentioned before, process mining can improve busi-

ness processes. However, the usefulness of mined

process models depends on, e.g., whether the model is

comprehensible for the stakeholder or not. It is inter-

esting to see how different abstraction levels inﬂuence

the appropriateness of the process model for a speciﬁc

stakeholder. We will determine a list of stakehold-

ers of a logistics organization and deﬁne the organi-

zational level of interest for each stakeholder. These

interest levels will form the basis of the abstraction

levels as deﬁned in phase 3 of our methodology.

To this end, multiple stakeholders inside logistics

organizations were considered. Such organizations

typically include both primary and secondary stake-

holders (Flod

en and Woxenius, 2021; Iacob et al.,

2019; Tolentino-Zondervan et al., 2021). Primary

stakeholders have a formal, or contractual relation-

ship to the organization, while secondary stakeholders

are not directly connected to the company (Gibson,

2000). Since secondary stakeholders are typically not

directly bothered with analyzing a process model, we

will not put them into our stakeholder list. Instead,

we will deﬁne primary stakeholders that represent the

needs of secondary stakeholders.

Stakeholders usually have different goals and in-

terests. Furthermore, stakeholders are generally con-

cerned with different kinds of information. It might

be that one particular stakeholder wants to know more

about the overall structure of the process, while an-

other stakeholder is more interested in speciﬁc (ab-

normal) activities. For each stakeholder, we will de-

ﬁne a general purpose that is aligned with his needs.

By considering the case study context, work of (An-

thony, 1965; Greenley and Foxall, 1997; Iacob et al.,

2019; Post et al., 2002) helped us to identify the goals

of the stakeholders. The list of stakeholders consid-

ered is shown in Table 2.

3.3 Establishing Abstraction Levels

The fuzzy miner is used for generating the process

models. This miner uses a combination of signiﬁ-

cance and correlation thresholds to simplify the re-

sulting model (Van der Aalst and Gunther, 2007;

unther and Van Der Aalst, 2007). Signiﬁcance is

about the relative importance of behavior, while cor-

relation is about the precedence relation of two events.

Being able to tune on these parameters, makes the

fuzzy miner a suitable candidate for handling com-

plex and unstructured real-life event logs.

The parameters of the fuzzy miner have differ-

ent inﬂuences on the obtained process model. Five

thresholds can help to simplify the process model.

Two of these thresholds, the preserve threshold and

the ratio threshold, will not be considered, because

these only have an inﬂuence on nodes with a conﬂict-

ing relation (G

unther and Van Der Aalst, 2007). Our

data set does not contain these types of relations. The

other three thresholds all inﬂuence the simpliﬁcation

in different ways. An overview of the thresholds can

be found in Table 3.

In general, the lower the deﬁned thresholds are,

the less abstract a process model will become. The

utility ratio, however, does not directly inﬂuence the

abstraction of the model since it focuses on a combi-

nation of signiﬁcance and correlation. Therefore, we

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

140

Table 2: An example of stakeholders with organizational abstraction levels.

Stakeholder Purpose Organizational interest

Operational board Identify the overall workﬂow of the company Top

CFO Get to know the overall cost picture and identifying the speciﬁc

causes of high costs

Top & Middle

Planner Identify all the steps an order undergoes and where delays occur Middle

Driver Find out what activities constitute to their speciﬁc task Bottom

Exception manager Spot exceptions and ﬁnd out how they occurred Bottom

IT expert Find out what parts of the process require more extensive logging Middle

Regulations expert Make sure all steps necessary for regulation measures are taken Bottom

Customer relations Ensure traceability and timeliness of the orders Bottom

Table 3: Considered thresholds for the fuzzy miner.

Threshold Value = 0 Value = 1 Application

Utility ratio (UR) High correlation/

low signiﬁcance

High signiﬁcance/

low correlation

Edge ﬁltering

Edge cutoff (EC) Diminishes utility ratio Ampliﬁes utility ratio Edge ﬁltering

Node cutoff (NC) Less abstract More abstract Node ﬁltering

will keep this value constant (to make sure it will not

bias our results). Due to the enrichment of the data,

there are not many activities with a high signiﬁcance.

Therefore, we kept the node cutoff at a relatively low

level, to prevent the model from containing only one

cluster.

In total, we consider four abstraction levels (A, B,

C, D) (see Table 4), based on discussions among the

authors of this paper. Abstraction level A is the most

abstract and, therefore, contains the least details. Ab-

straction level D is the least abstract. We consider

model A to correspond to the top organizational level,

models B and C with the middle organization level,

and model D with the bottom level.

Table 4: Abstraction levels.

Abstraction level UR EC NC

A 0.5 1.0 0.4

B 0.5 0.8 0.25

C 0.5 0.6 0.1

D 0.5 0.4 0.0

3.4 Deﬁning Quality Measurements

Fitness is an important quantitative measure in pro-

cess mining, which indicates how well the behavior

as described in an event log is displayed in the pro-

cess model (Van der Aalst et al., 2006). Fitness is

useful for getting an idea of the quality of the process

model. If we know that such a model is not properly

representing reality, we could also be less interested in

some other details of the process model. Apart from

the ﬁtness, several statistics of the process model are

used to obtain an understanding of the complexity of

the process models. The ﬁrst statistic is the level of

detail (a percentage that displays how many nodes are

preserved in the model). The other statistics include

the number of nodes, edges, and clusters shown in the

model.

Besides these quantitative measures, we also in-

clude a qualitative assessment by using the TAM in

combination with an expert analysis. The analysis

consists of a panel of stakeholders/experts that evalu-

ate the process models, while reasoning from the per-

spective of a certain stakeholder. TAM offers a set of

questions about the perceived usefulness and ease of

use for an end-user. TAM is a well-known approach

for measuring the acceptance of new technology, and

has already been adopted by researchers within the

context of process mining projects (Wynn et al., 2017;

Graafmans et al., 2021). We will adapt the original

TAM in order to encapsulate the views of stakehold-

ers differ per abstraction level. This adaption makes

the TAM suitable for our research to reason on the

quality of the process models by directly taking into

account the opinions of relevant stakeholders, as de-

ﬁned in Section 3.2. Besides our own research, the

proposed TAM can be used in other process mining

research, that actively involves stakeholder opinions.

These quality measurements will be used to rea-

son on the quality of a process model. We deﬁne qual-

ity as the appropriateness of a process model in terms

of its ﬁtness, perceived usefulness, and perceived ease

of use.

3.5 Modeling

The modeling phase is where the process models will

be speciﬁed at the abstraction levels (as deﬁned in

Section 3.3). The resulting process models are shown

in Figure 2.

A Methodology for Aligning Process Model Abstraction Levels and Stakeholder Needs

141

(a) Process model A (b) Process model B

Figure 2: Process model abstractions.

3.6 Evaluation

The ﬁnal phase of our methodology is about the eval-

uation of the generated process models. The goal of

the evaluation is to deﬁne which level of abstraction is

the most useful for a particular stakeholder. By using

the measurements deﬁned in phase 4 (Section 3.4) we

have performed both quantitative and qualitative eval-

uation.

3.6.1 Quantitative Results

Our quantitative analysis (see Table 5) shows that,

once we get to the more abstract levels, fewer nodes

are present and more clusters exist. Model D, the

most detailed model, includes all activities and many

edges. This causes the ﬁtness of the model to be

high, since many speciﬁc traces are visible in model

D. Although these speciﬁc traces can give interesting

insights, it might be the case that the model is per-

ceived as cluttered. Model C already places certain

nodes inside clusters and heavily reduces the number

of edges. Unfortunately, the model has a low ﬁtness

score. An explanation can be that many edges have

been removed and that not that many nodes are placed

in clusters. Model B puts even more nodes in clusters

and reduces the number of edges even further. The

fact that the clusters are relatively small might im-

prove comprehensibility, since large clusters can be

perceived as a black box. Although its ﬁtness is not as

high as models A and D, we still consider the ﬁtness to

be of a sufﬁcient level. Finally, model A provides the

least detail by removing quite some edges and putting

almost all nodes inside clusters. This causes the ﬁt-

ness to be high, however, it might be the case that the

model is perceived as incomprehensible due to this

lack of detail.

3.6.2 Qualitative Results

The results of the expert/stakeholder analysis in com-

bination with the TAM can be seen in Table 6. The

table shows how each model performs in terms of

usefulness and ease of use, as perceived by the dif-

ferent stakeholders. In total, eight domain experts

were consulted for the expert analysis. The experts

included mainly academics active in the logistics do-

main. Every expert reasoned from the perspective of

two stakeholders. This means that every stakeholder

is reviewed twice, by two different experts. Each ex-

pert gives a score ranging from 1 to 5 on a process

model, while reasoning from the perspective of a par-

ticular stakeholder. Hence, every process model was

reviewed sixteen times, from the view of eight stake-

holders in total.

When observing the results (as summarized in Ta-

ble 7), we see that model A scores worst in terms of

usefulness. As mentioned in the quantitative analysis

part, due to the lack of detail, the model does proba-

bly not provide relevant information. The other mod-

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

142

Table 5: Evaluation of process models.

Model Fitness Detail Nodes Edges Clusters (nodes)

A 99.84% 52.44% 13 21 3(28, 24, 8)

B 96.26% 69.17% 20 40 5(24, 12, 8, 5, 3)

C 91.99% 88.71% 38 93 3(19, 6, 2)

D 99.00% 100% 73 150+ 0

Table 6: Results Technology Acceptance Model (scale 1-5).

Construct Average

General Information

1. I have much experience with business

process modelling in general.

3.19

2. I have much experience with process

mining in general.

2.63

Model A B C D

Usefulness

3. The information presented in this model

is useful for my daily job.

1.94 3.13 3.56 3.19

4. The model is suitable for gaining new

insights about the business process.

1.56 3.31 3.81 3.50

5. The model contains detailed information

about the business process.

1.38 2.75 4.13 4.69

6. The model helps forming an understanding

of the business process in general.

2.31 3.88 3.56 2.25

Average 1.80 3.27 3.77 3.41

Ease of Use

7. The model is understandable when taking

a ﬁrst look at it.

4.00 3.56 2.88 1.56

8. It is easy to learn understanding this

model.

4.06 3.50 3.06 1.75

9. It is easy to explain this model to other

persons inside the organization.

3.75 3.69 3.06 1.50

10. Someone without experience in process

mining is able to understand this model.

3.50 3.44 2.56 1.50

11. I will use the information obtained from

this model in my daily job.

1.88 3.31 3.63 2.63

12. This model helps me achieve my purpose

inside the organization.

1.69 3.31 3.81 2.94

Average 3.15 3.47 3.17 1.98

els all score above 3, with model C scoring the high-

est. The combination of abstracting speciﬁc behavior,

whilst still providing enough detail, showed to be use-

ful for the stakeholders.

As for the ease of use, the opinion of the stake-

holders on models A and D has entirely changed. The

fact that model D provides more information about

the overall process, can make it challenging to un-

derstand for the stakeholders, which may result in a

low score for ease of use. Models B and C still score

consistently above 3, indicating that stakeholders also

perceive these models as being (relatively) compre-

hensible and easy to use.

Using our understanding of the perceived useful-

ness, the perceived ease of use, and the quantita-

tive metrics of every process model, an alignment of

stakeholders and process model complexity can be

made. The way in which the measurements of this

alignment are used is situation dependent and one

may favor one construct more than the other. Factors

such as the type of stakeholder and the organizational

context inﬂuence the preferred prioritization of each

measurement. By excluding model C (e.g., because of

its low ﬁtness) and having discussions about the score

on perceived usefulness and ease of use, authors of

the present work arrived at the alignment made in Ta-

ble 7. Our methodology allows to choose an approach

in evaluating the constructs, depending on the con-

text in which they apply our methodology. Deﬁning a

decision-making method (e.g., a ranking mechanism

or any, more formal, form of multi-criteria decision-

making method) is something that can be introduced

in a future extension of the methodology.

Table 7: An example of stakeholders with preferable ab-

straction levels.

Stakeholder Abstraction level Stakeholder Abstraction level

Operational board B Exception manager D

CFO B IT expert B

Planner B Regulations expert D

Driver B Customer relations B

Notice that we involved experts that reasoned

from the perspectives of stakeholders, instead of a

(group of) representative stakeholder(s). In princi-

ple, our proposed methodology could be applied to

both experts and stakeholders. We decided to involve

experts, as they were familiar with the semantics of

process models and the concerned case study. Yet,

not all experts were experienced with the stakeholder

roles. However, we justify this choice by (1) using

more than one expert reasoning on each stakeholder,

(2) providing the experts a functional description of

the stakeholders (Table 2), and (3) restating that our

purpose concerns demonstrating a design artifact as

part of a design science cycle instead of exhaustively

evaluating a case study.

4 RELATED WORK

Events can be recorded at a (very) granular level, and

if not dealt with appropriately (e.g., through process

mining discovery algorithms) this can result in incom-

prehensible process models. What an “appropriate”

level of detail is needed is debatable. Generated pro-

cess models should be understandable (Van Cruchten

and Weigand, 2018a). Mostly it is assumed that event

data are of the same and bear an appropriate level

of granularity. However, in reality, granular event

logs often produce either spaghetti or lasagne pro-

cess models (Van der Aalst, 2016). To understand

A Methodology for Aligning Process Model Abstraction Levels and Stakeholder Needs

143

these complex and (semi-)unstructured models there

are many approaches proposed in literature. To men-

tion some, there are pre-processing techniques that al-

low an appropriate level of granularity as for example

identiﬁed by (Van Zelst et al., 2021) and approaches

that use the complex process model and choose the

level of details based on who is expected to beneﬁt

from the process model. The latter one can be exhib-

ited in particular views for, e.g., customers (Bernard

and Andritsos, 2018) or healthcare providers (Mans

et al., 2012).

There exist many algorithms that focus on event

log abstraction within the process mining discipline.

For example, (De Medeiros et al., 2007) proposed a

clustering algorithm for reducing the level of abstrac-

tion in process models. This algorithm iteratively

makes clusters based on the event log, until the ob-

tained process models do not over-generalize certain

activities of the event log. (Becker and Intoyoad,

2017) explores how a k-medoids algorithm can be

used to cluster heterogeneous datasets. They check

the characteristics of the obtained models for differ-

ent levels (of k). Their results indicate that it is useful

to evaluate the resulting process models based on a

speciﬁc purpose. However, a common value of k that

is appropriate for several stakeholder purposes is not

found. (Dos Santos Garcia et al., 2019) attempts to

address this problem but also calls for future research

on methods to determine the main processes for par-

ticular purposes (e.g., stakeholders). (Baier et al.,

2014) uses an abstraction approach based on external

domain knowledge. They stressed the importance of

making process models understandable for business

users, by working with an appropriate level of abstrac-

tion (Van der Aalst and Gunther, 2007; G

unther and

Van Der Aalst, 2007). (Fazzinga et al., 2018a) de-

scribe how event logs with low-level events that seem

to have no reference to high-level activities, can be

transformed to the preferred abstraction level of an

analyst. Although their method helps with adjusting

the event log such that it is more suitable for creat-

ing a process model, it does not speciﬁcally address

how different abstraction levels inﬂuence the appro-

priateness of the process model for a given stake-

holder. Other work from (Fazzinga et al., 2018c) pro-

poses a framework that induces process models to de-

scribe the process at an activity level, to better suit

the needs of process model analysts. Although this

framework actively seeks to shift a process model to-

wards an abstraction level that is more appropriate for

model analysts, it does not elaborate on how distinct

stakeholders require different levels of abstraction. In

our study, we take explicitly into account the perspec-

tive of different stakeholders when assessing process

models. In line with this, our work also relates to

a client-server-based application proposed by (Yazdi

et al., 2021) to gradually abstract ﬁne-grained event

logs to higher levels without losing essential informa-

tion, thereby enabling the domain experts to use the

appropriate process model for further analysis.

Our work also taps into a recent discussion on

the use of agent-based modeling in combination with

(data-driven) process mining techniques. The deci-

sion logic of the case study considered in the present

paper uses agent-based modeling (Bemthuis et al.,

2020), which acts as a natural recourse when incorpo-

rating human-interpretation capabilities or when in-

teracting with humans. In our work, we give a qualita-

tive and quantitative assessment of the resulting event

logs which are originated from complex agent inter-

actions. Thereby, we also put forward the call for re-

search on addressing how agent-based models could

affect the quality of process mined models (Bemthuis

et al., 2019).

5 CONCLUSIONS AND FUTURE

WORK

The relevance and comprehensibility of a process

model for a particular stakeholder are greatly inﬂu-

enced by the level of abstraction of the process model.

Yet, providing a suitable process model for a desig-

nated stakeholder can be challenging. Current liter-

ature on stakeholder analysis and process model ab-

straction in the process mining discipline acknowl-

edges the need for aligning a stakeholders’ purposes

to relevant process models. Therefore, we proposed a

methodology to align stakeholder needs with process

model complexity. This methodology consists of six

phases that each contribute to aligning the complexity

of generated process models to the needs of particu-

lar stakeholders. A logistics case study, in which we

used the TAM as a qualitative measurement, demon-

strates the usefulness of our methodology for properly

abstracting process models while taking into account

the needs of stakeholders.

Some limitations and potential improvements in

this study are as follows. The ﬁrst is related to our

case study, which involves only a few complex traces

and the resulting process models were not exhaustive.

Although we enriched the event logs by adding an ad-

ditional attribute and considering multiple stakehold-

ers, further work relies on more realistic and complex

case studies to validate our methodology. Case stud-

ies in diverse domains and with representative groups

of experts/stakeholders are desired to validate the pre-

sented methodology.

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

144

Second, the use of both quantitative and qualita-

tive metrics to assess the quality of a process model

from the perspective of different stakeholders is only

explored to a limited extent. Future work could focus

on how the interaction between a stakeholder/expert

and mined process models will occur more speciﬁ-

cally. Within the logistics domain, a suggestion could

be to focus on process mining use cases involving the

Open Trip Model (OTM). A conceptual mapping al-

ready outlined that OTM provides a promising way to

unify storage, integration, interoperability, and query-

ing of logistics event data (Piest et al., 2021).

Third, the application of TAM to reason on pro-

cess model abstractions requires a deeper investiga-

tion of the role and expertise of experts/stakeholders.

Further improvement lies in exploring possible ways

to incorporate not only domain knowledge but also

technical knowledge of the assessors. Our case study

involved experts that reasoned on behalf of stakehold-

ers, while our methodology may also be applicable to

actual stakeholders in the ﬁeld. The notion that the

same identical process may have different representa-

tions for different stakeholders is an interesting out-

come. We call for further work on the use of abstrac-

tion in process mining not only based on topology but

also on content and the operational relevance of indi-

vidual process steps. Hence, the graphical layout of

a model should be separated from, or at least com-

plementary to, its (knowledge) content. Recent de-

velopments on domain-knowledge-utilizing process

discovery algorithms underscores this need (Schuster

et al., 2022).

Fourth, it may also be fruitful to investigate how

differentiation by means of abstraction levels (tailored

for speciﬁc stakeholders) can be applied to other parts

of a process mining project. For example, during the

data processing stage or when evaluating a process

mining project. Algorithms and methodologies could

explicitly take into account multi-stakeholder views.

ACKNOWLEDGEMENTS

This work is part of the project DataRel (grant

628.009.015) which is (partly) ﬁnanced by the Dutch

Research Council (NWO). The authors would like to

thank all project partners as well as Syeda Sohail for

their support. The authors also thank the anonymous

reviewers for their constructive feedback.

REFERENCES

Anthony, R. N. (1965). Planning and control systems: a

framework for analysis. Division of Research, Gradu-

ate School of Business Administration, Harvard.

Baier, T., Mendling, J., and Weske, M. (2014). Bridging

abstraction layers in process mining. Information Sys-

tems, 46:123–139.

Becker, T. and Intoyoad, W. (2017). Context aware process

mining in logistics. Procedia CIRP, 63:557–562.

Bemthuis, R., Mes, M., Iacob, M., and Havinga, P.

(2021a). Data underlying the paper: Using agent-

based simulation for emergent behavior detection in

cyber-physical systems. DOI: 10.4121/14743263.

4TU.ResearchData.

Bemthuis, R., Mes, M., Iacob, M.-E., and Havinga, P.

(2020). Using agent-based simulation for emergent

behavior detection in cyber-physical systems. In 2020

Winter Simulation Conference (WSC), pages 230–241.

IEEE.

Bemthuis, R., van Slooten, N., Arachchige, J. J., and Piest,

J. P. S. (2021b). A classiﬁcation of process min-

ing bottleneck analysis techniques for operational sup-

port. In Proceedings of the 18th International Confer-

ence on e-Business, pages 127–135. SCITEPRESS.

Bemthuis, R. H., Koot, M., Mes, M. R. K., Bukhsh, F. A.,

Iacob, M.-E., and Meratnia, N. (2019). An agent-

based process mining architecture for emergent be-

havior analysis. In 2019 IEEE 23rd International

Enterprise Distributed Object Computing Workshop

(EDOCW), pages 54–64. IEEE.

Bernard, G. and Andritsos, P. (2018). CJM-ab: Abstracting

customer journey maps using process mining. In In-

ternational Conference on Advanced Information Sys-

tems Engineering, pages 49–56. Springer.

Davis, F. D. (1989). Perceived usefulness, perceived ease of

use, and user acceptance of information technology.

MIS quarterly, pages 319–340.

De Medeiros, A. K. A., Guzzo, A., Greco, G., Van der Aalst,

W. M. P., Weijters, A. J. M. M., Van Dongen, B. F.,

and Sacca, D. (2007). Process mining based on clus-

tering: A quest for precision. In International confer-

ence on business process management, pages 17–29.

Springer.

Diba, K., Batoulis, K., Weidlich, M., and Weske, M. (2020).

Extraction, correlation, and abstraction of event data

for process mining. Wiley Interdisciplinary Reviews:

Data Mining and Knowledge Discovery, 10(3):e1346.

Dos Santos Garcia, C., Meincheim, A., Junior, E. R. F., Dal-

lagassa, M. R., Sato, D. M. V., Carvalho, D. R., San-

tos, E. A. P., and Scalabrin, E. E. (2019). Process min-

ing techniques and applications–a systematic mapping

study. Expert Systems with Applications, 133:260–

295.

Fazzinga, B., Flesca, S., Furfaro, F., Masciari, E., and Pon-

tieri, L. (2018a). Efﬁciently interpreting traces of low

level events in business process logs. Information Sys-

tems, 73:1–24.

Fazzinga, B., Flesca, S., Furfaro, F., and Pontieri, L.

(2018b). Online and ofﬂine classiﬁcation of traces of

A Methodology for Aligning Process Model Abstraction Levels and Stakeholder Needs

145

event logs on the basis of security risks. Journal of

Intelligent Information Systems, 50(1):195–230.

Fazzinga, B., Flesca, S., Furfaro, F., and Pontieri, L.

(2018c). Process discovery from low-level event logs.

In International Conference on Advanced Information

Systems Engineering, pages 257–273. Springer.

Flod

en, J. and Woxenius, J. (2021). A stakeholder analysis

of actors and networks for land transport of dangerous

goods. Research in Transportation Business & Man-

agement, page 100629.

Gibson, K. (2000). The moral basis of stakeholder theory.

Journal of business ethics, pages 245–257.

Graafmans, T., Turetken, O., Poppelaars, H., and Fahland,

D. (2021). Process mining for six sigma. Business &

Information Systems Engineering, 63(3):277–300.

Greenley, G. E. and Foxall, G. R. (1997). Multiple stake-

holder orientation in uk companies and the implica-

tions for company performance. Journal of Manage-

ment Studies, 34(2):259–284.

Gruhn, V. and Laue, R. (2007). Approaches for busi-

ness process model complexity metrics. In Technolo-

gies for business information systems, pages 13–24.

Springer.

unther, C. W. and Van Der Aalst, W. M. P. (2007).

Fuzzy mining–adaptive process simpliﬁcation based

on multi-perspective metrics. In International con-

ference on business process management, pages 328–

343. Springer.

Iacob, M.-E., Charismadiptya, G., van Sinderen, M., and

Piest, J. P. S. (2019). An architecture for situation-

aware smart logistics. In 2019 IEEE 23rd Inter-

national Enterprise Distributed Object Computing

Workshop (EDOCW), pages 108–117. IEEE.

Iacob, M. E., Monteban, J., Van Sinderen, M., Hegeman,

E., and Bitaraf, K. (2018). Measuring enterprise archi-

tecture complexity. In 2018 IEEE 22nd International

Enterprise Distributed Object Computing Workshop

(EDOCW), pages 115–124. IEEE.

Kumar, M. V. M., Thomas, L., and Annappa, B. (2017).

Distilling lasagna from spaghetti processes. In Pro-

ceedings of the 2017 International Conference on In-

telligent Systems, Metaheuristics & Swarm Intelli-

gence, pages 157–161.

Leemans, S. J. J., Goel, K., and van Zelst, S. J. (2020).

Using multi-level information in hierarchical process

mining: Balancing behavioural quality and model

complexity. In 2020 2nd International Conference on

Process Mining (ICPM), pages 137–144. IEEE.

Mans, R. S., Van der Aalst, W. M. P., Vanwersch, R. J. B.,

and Moleman, A. J. (2012). Process mining in health-

care: Data challenges when answering frequently

posed questions. In Process Support and Knowl-

edge Representation in Health Care, pages 140–153.

Springer.

Muketha, G. M., Abd Ghani, A. A., Selamat, M. H., and

Atan, R. (2010). A survey of business processes

complexity metrics. Information Technology Journal,

9(7):1336–1344.

Peffers, K., Tuunanen, T., Rothenberger, M. A., and Chat-

terjee, S. (2007). A design science research method-

ology for information systems research. Journal of

management information systems, 24(3):45–77.

Piest, J. P. S., Cutinha, J. A., Bemthuis, R. H., and Bukhsh,

F. A. (2021). Evaluating the use of the open trip model

for process mining: An informal conceptual mapping

study in logistics. In Proceedings of the 23rd Interna-

tional Conference on Enterprise Information Systems,

pages 290–296. SCITEPRESS.

Post, J. E., Preston, L. E., and Sachs, S. (2002). Managing

the extended enterprise: The new stakeholder view.

California management review, 45(1):6–28.

Schuster, D., van Zelst, S. J., and van der Aalst, W. M. P.

(2022). Utilizing domain knowledge in data-driven

process discovery: A literature review. Computers in

Industry, 137:103612.

Tolentino-Zondervan, F., Bogers, E., and van de Sande, L.

(2021). A Managerial and Behavioral Approach in

Aligning Stakeholder Goals in Sustainable Last Mile

Logistics: A Case Study in the Netherlands. Sustain-

ability, 13(8).

Van Cruchten, R. M. E. R. and Weigand, H. H. (2018a).

Process mining in logistics: The need for rule-based

data abstraction. In 2018 12th International Confer-

ence on Research Challenges in Information Science

(RCIS), pages 1–9. IEEE.

Van Cruchten, R. R. and Weigand, H. H. (2018b). Process

mining in logistics: The need for rule-based data ab-

straction. In 2018 12th International Conference on

Research Challenges in Information Science (RCIS),

pages 1–9. IEEE.

Van der Aalst, W. M. P. (2016). Process mining - Data

science in action. Springer.

Van der Aalst, W. M. P., De Medeiros, A. K. A., and Wei-

jters, A. J. M. M. (2006). Process equivalence: Com-

paring two process models based on observed behav-

ior. In International conference on business process

management, pages 129–144. Springer.

Van der Aalst, W. M. P. and Gunther, C. W. (2007). Finding

structure in unstructured processes: The case for pro-

cess mining. In Seventh International Conference on

Application of Concurrency to System Design (ACSD

2007), pages 3–12. IEEE.

Van Eck, M. L., Lu, X., Leemans, S. J. J., and Van der Aalst,

W. M. P. (2015). PM

: A process mining project

methodology. In International Conference on Ad-

vanced Information Systems Engineering, pages 297–

313. Springer.

Van Zelst, S. J., Mannhardt, F., de Leoni, M., and

Koschmider, A. (2021). Event abstraction in process

mining: literature review and taxonomy. Granular

Computing, 6(3):719–736.

Wirth, R. and Hipp, J. (2000). CRISP-DM: Towards a stan-

dard process model for data mining. In Proceedings of

the 4th international conference on the practical ap-

plications of knowledge discovery and data mining.

Wynn, M. T., Poppe, E., Xu, J., ter Hofstede, A. H., Brown,

R., Pini, A., and Van der Aalst, W. M. P. (2017).

Processproﬁler3d: A visualisation framework for log-

based process performance comparison. Decision

Support Systems, 100:93–108.

ICEIS 2022 - 24th International Conference on Enterprise Information Systems

146

Yazdi, M. A., Ghalatia, P. F., and Heinrichs, B. (2021).

Event log abstraction in client-server applications. In

Proceedings of the 13th International Joint Confer-

ence on Knowledge Discovery, Knowledge Engineer-

ing and Knowledge Management - Volume 1: KDIR,,

pages 27–36. SCITEPRESS.

Zakarija, I.,

Skopljanac-Ma

cina, F., and Bla

skovi

c, B.

(2015). Discovering process model from incomplete

log using process mining. In 2015 57th International

Symposium ELMAR, pages 117–120. IEEE.

A Methodology for Aligning Process Model Abstraction Levels and Stakeholder Needs

147