Understandability Issues of Approaches
Supporting Business Process Variability?
Victoria Torres1, Stefan Zugal2, Barbara Weber2, Manfred Reichert3,
Clara Ayora1, and Vicente Pelechano1
1Universitat Polit`ecnica de Val`encia, Spain
{vtorres,cayora,pele}@pros.upv.es
2University of Innsbruck, Austria
{stefan.zugal,barbara.weber}@uibk.ac.at
3University of Ulm, Germany
manfred.reichert@uni-ulm.de
Abstract. The increasing adoption of Process-Aware Information Sys-
tems, together with the reuse of process knowledge, has led to the emer-
gence of process model repositories with large process families, i.e., col-
lections of related process model variants. For managing such related
model collections two types of approaches exist. While behavioral ap-
proaches take supersets of variants and derive a process variant by hiding
and blocking process elements, structural approaches take a base process
model as input and derive a process variant by applying a set of change
operations to it. However, at the current stage no framework for assess-
ing these approaches exists and it is not yet clear which approach should
be better used and under which circumstances. Therefore, to give first
insights about this issue, this work compares both approaches in terms
of understandability of the produced process model artifacts, which is
fundamental for the management of process families and the reuse of
their contained process fragments. In addition, the comparison can serve
as theoretical basis for conducting experiments as well as for fostering
the development of tools managing business process variability.
1 Introduction
The increasing adoption of Process-Aware Information Systems (PAISs) by in-
dustry has led to large collections of process models in a variety of application
domains. Despite the particularities found in specific domains, many of these
models share common parts of their definition (e.g., activities). We denote such
related process variant models as process family in the following. By using com-
mon business process modeling languages such as BPMN, EPC, YAWL, or UML
Activity Diagrams, process families can just be represented either in separate
process models (one per process variant) or in a unified model using conditional
?This work has been developed with the support of MICINN under the project EV-
ERYWARE TIN2010-18011.
2 Victoria Torres et al.
branching for describing differences and subprocesses for capturing common-
alities [1]. Even though the use of subprocesses improves the reuse of process
fragments within the process family, this is not sufficient to capture variabil-
ity. In particular, dependencies between process fragments as well as context
information cannot be represented explicitly in the model, aggravating model
maintenance.
To properly handle large process families (i.e., avoid redundancies, foster
reusability, and reduce modeling efforts) several proposals have been developed
in recent years (e.g., Provop [1], C-EPC [2], PESOA [3], Rule representation
and processing [4], and PPM [5]), which can be classified as either behavioral
or structural approaches. Behavioral approaches represent all members of the
family within the same model artifact, capturing both, commonalities and par-
ticularities of all process variants. In turn, structural approaches use different
artifacts to represent the family, e.g., by using a base model to which structural
changes such as inserting, deleting, or moving activities may be applied to derive
process variants.
To foster reusability of process model families, understandability of the cre-
ated artifacts is essential. So far, however, no experimental insights regarding
quality aspects (e.g., understandability) are available and it is not clear under
which circumstances the use of one approach is more appropriate than the other.
Since the results of assessing a process model’s understandability significantly
depend on the specific understandability tasks [6, 7], we have structured the
comparison of both approaches along several such tasks using a process family
from the film industry. For each of the tasks, the paper discusses the process
followed by a model reader to accomplish such tasks using both approaches. In
addition, we use cognitive psychology as a tool for explaining the differences be-
tween the two approaches. This comparison will provides us the theoretical basis
for conducting experiments as well as for fostering the development of tools for
managing variability in business processes.
The remaining sections are structured as follows. Section 2 presents a pro-
cess family from the film industry along which the two different approaches are
compared. Section 3 provides basic notions and introduces the behavioral and
structural approaches for dealing with variability in business processes. Then,
Section 4 describes the concepts from cognitive psychology that are used to
assess understandability aspects of both approaches. Based on these concepts,
Section 5 provides a qualitative comparison for a set of comprehension tasks re-
garding the understandability of the modeling artifacts built by each approach.
Section 6 then presents an overview of related work. Finally, Section 7 concludes
the paper and gives an outlook.
2 Example of a Process Family
As running example, we consider a modified version of a process family from the
film industry for editing a screen project, which varies depending on the shooting
media and delivery media used [8]. First, footage is received, either in tape,film,
Qualitative Comparison 3
or tape and film and prepared for edition depending on the shooting medium.
Then offline edition is performed. Next, the cut stage is performed through
online edit (if the shooting medium is tape), through negmatching (in the case
of film), or through both cuts (when shooting media is tape and film). After this
point, the finishing on a delivery medium phase starts. For this purpose, the
delivery media (e.g., tape, film, tape and film, or new medium) must be chosen.
As a result, different variants exist. For example, when shooting media is film
or tape and film and cutting has been performed through negmatching, a finish
on film has to be performed to maintain the quality of the delivery medium.
On the contrary, if the cutting is performed through online editing and film
or tape and film is the delivery medium, record digital film master needs to be
mandatorily performed to transfer the editing results to film. Similarly, telecine
transfer will be performed only if negmatching is performed previously and the
expected delivery format is tape or new medium. Finally, if neither tape or film
have been chosen activity finish on new medium must be performed.
3 Approaches for Modeling BP Variability
To properly represent variability in a BP model, it is important to define (1)
what parts of the BP model may vary according to a specific context, (2) what
alternatives exist in each of those parts, and (3) which conditions make these
alternatives being selected. The first issue refers to the precise identification of
the parts being subject to variation, which are commonly known as variation
points (e.g., the edition phase which depends on the type of medium used for
footage—tape,film, or tape and film). The second issue refers to the different
alternatives that exist for all those variation points (e.g., the alternative activi-
ties for the edition phase are edit footage online (when shooting medium is tape)
and perform negmatching (when shooting medium is film)). In addition, some
models may require the definition of relationships (e.g., inclusion, exclusion) be-
tween alternative process fragment from different variation points (e.g., activity
record digital film master has to be performed when negmatching has been per-
formed and the delivery medium is film or tape and film). The third issue refers
to the context of these variations, which is usually represented by a set of vari-
ables gathered in a context model in which the BP model is used (e.g., shooting
and delivery media types). These issues are not addressed by common business
process modeling languages (e.g., BPMN, EPC, YAWL) and their consideration
turns crucial to properly represent and manage process families. The following
subsections present two different approaches targeted at the representation of
such families, i.e., behavioural and structural.
3.1 Behavioural Approach
The behavioral approach represents a process family in a single artifact, known as
configurable process model capturing both the commonalities and particularities
4 Victoria Torres et al.
of the process variants reflecting all possible behavior. Proposals that have been
developed following this approach include C-EPC [2], C-YAWL [9] or Feature-
EPC [10]. In the following we take C-EPC [2] as the representative proposal
since it constitutes the most well-known and mostly cited proposal in literature.
C-EPC extends EPC with configurable elements (i.e., configurable nodes and
configuration requirements) to explicitly model variability. Fig. 1 illustrates the
configurable process model representing the postproduction process (cf. Section
2). On the one hand, configurable nodes (i.e., functions and connectors) are rep-
resented graphically with thick solid borders and define variations points in the
model where different alternatives may exist. Specifically, configurable functions
(e.g., activity telecine transfer in Fig. 1) can be configured as ON (i.e., function
is kept in the model), OFF (i.e., function is removed from the model), or OPT
(i.e., conditional branching is included in the model deferring the decision to
run-time). Configurable connectors, in turn, can be configured to an equally or
more restrictive connector. For example, a configurable OR can be configured as
a regular OR (not applying any restrictions), or can restrict its behaviour by con-
figuring it as an XOR (i.e., selecting one of the outgoing/incoming alternatives),
AND (i.e., selecting all of the outgoing/incoming alternatives), or SEQn(i.e.,
reducing the alternatives for the configuration of the connector to just one of its
outgoing/incoming sequences). Table 1 summarizes all possible configurations
for the different types of connectors (default configurations, i.e., configurations
applied when no constraints are defined, are marked with asterisks).
Config. connector OR XOR AND SEQn
OR X* X X X
XOR X* X
AND X
Table 1. Allowed configurations of configurable connectors
On the other hand, configuration requirements are graphically represented
as tags attached to configurable nodes and formalize, by means of logical pred-
icates, domain constraints related to the attached nodes. The configuration of
the node will the be made based on the evaluation of the attached configuration
requirements. For example, requirement 5 states that configurable connector 3
has to be configured as SEQ3awhen configurable connector 1 has been con-
figured as SEQ1a, i.e., activity edit footage online has to be performed when
shooting medium is tape, forbidding implicitly the execution of activity per-
form negmatching). However, configurable nodes not always have requirements
attached to them (since they are only included when there is a constraint regard-
ing their configuration). When no requirements are attached to a configurable
node, this should be transformed into a regular one, maintaining the behaviour
of the original connector and deferring the configuration decision to run-time.
Qualitative Comparison 5
Note that some requirements (i.e., req. 1–3 and 9–12) comprise context vari-
ables, providing information regarding the context that drives the configuration
of the associated configurable nodes, while others (i.e., req. 4–8 and 13–16) are
expressed in terms of the structure of the model only. In the original proposal
[8] only requirements of the latter type are used and the context variables only
become explicit when using the proposal in combination with the questionnaire
approach [32]. To make the C-EPC model better understandable we added re-
quirements of the former category to make the context for choosing a certain
configuration explicit.
Prepare
tape for edit
V
V
V
V
V
Tape shoot
finished Film shoot
finished
Footage
prepared for
edit
Edit
finished
Post-
production
finished
Requirement 16
(OR5= ‘SEQ5a’
Finish on tape= ‘OFF’) ->
Finish on new medium = ‘ON’
SEQ2a SEQ2b
2
SEQ3a SEQ3b
3
4
SEQ5a SEQ5b
5
6
Requirement 13
Record digital film master = ‘ON’ <->
(OR3= ‘SEQ3a’
(OR5= ‘AND’ OR5= ‘SEQ5b’))
Requirement 14
Telecine transfer = ‘ON’ <->
((OR3= ‘SEQ3b’ OR3= ‚AND‘)
(Finish on tape = ‘ON’
Finish on new medium = ‘ON’))
Requirement 9
delivery_media = ‘FILM’ ->
(OR5= ‘SEQ5b’)
Requirement 11
Prepare film
for edit
Edit footage
offline
Edit footage
online Perform
negmatching
Finish on
film
Record
digital film
master
Finish on
new
medium
Receive
footage
Shooting
finished
V
Requirement 4
OR2= OR1
1
SEQ1a SEQ1b
SEQ4a SEQ4b
Requirement 15
OR6= OR5
Requirement 8
OR4= OR3
Configurable
function
Configurable
OR connector
Configuration
requirement
V
Function
Event
Requirement 2
shooting_media = ‘FILM’ ->
(OR1= ‘SEQ1b’)
Requirement 1
shooting_media = ‘TAPE’
-> (OR1= ‘SEQ1a’)
SEQ6a SEQ6b
Finish on
tape
Telecine
transfer
Requirement 3
shooting_media = ‘TAPE FILM’
-> (OR1= ‘AND’)
delivery_media = ‘NEW MEDIUM’
-> (OR5= ‘SEQ5a’
Finish on tape = ‘OFF‘)
Requirement 12
Requirement 10
delivery_media = ‘TAPE FILM’ ->
(OR5= ‘AND’
Finish on tape = ‘ON‘)
delivery_media = ‘TAPE’ ->
(OR5= ‘SEQ5a’
Finish on tape = ‘ON‘)
Requirement 5
OR1= ‘SEQ1a’ ->
(OR3= ‘SEQ3a’)
Requirement 7
OR1= ‘AND’ ->
(OR3= ‘AND’)
Requirement 6
OR1= ‘SEQ1b’ ->
(OR3= ‘SEQ3b’)
Fig. 1. C-EPC model for the screen postproduction process
6 Victoria Torres et al.
3.2 Structural Approach
This approach proposes a gradual construction of the process family by modify-
ing the structure of a specific process variant (called base model) at specific points
(i.e., variation points) through change operations. Following this approach, we
find proposals such as Provop [1] or Rule representation and processing [4]. In
the following we use Provop as representative for the structural approach, since
it can be considered the most widely used proposal for this approach. Fig. 2
illustrates the Provop proposal using the screen postproduction process (cf. Sec-
tion 2). Provop allows creating process variants by adjusting the base model (cf.
top part of Fig. 2) by the application of a set of high-level change operations
between a couple of adjustment points. Depending on the modeling situation
and on the existing process landscape, the construction of the base model can
differ. [11] presents different policies (i.e., standard process, most frequently used
process, minimal average distance, superset of all process variants, and intersec-
tion of all process variants) to perform such construction. Furthermore, Provop
allows for more complex configuration adjustments by grouping multiple change
operations into so-called change options (e.g., option 1 combines 2 delete and
2 insert operations). Change options are associated with a context rule, which
is used at configuration time to decide, whether a certain option is applicable
for the given context (e.g., regarding option 1 the context rule states that this
option should only be applied if shooting media is tape). In addition, Provop
allows for an explicit representation of different dimensions of the context (i.e.,
context variables and allowed values) by means of the context model (cf. bot-
tom left part of Fig. 2). Finally, to prevent the derivation of semantically invalid
variants, Provop provides the constraint model (cf. bottom right part of Fig. 2)
which allows defining inclusion,exclusion,order of application,hierarchy, and
cardinality relationships between change options (e.g., the application of option
1excludes the application of option 2).
4 Concepts from Cognitive Psychology
In order to discuss differences between C-EPC and Provop in Section 5, we will
make use the concepts of external memory,abstraction, and split-attention effect
from cognitive psychology as introduced in the following and transfer them to
the domain of BP variability.
Basically, three different problem-solving “programs” or “processes” are
known from cognitive psychology: search, recognition, and inference [12]. Search
and recognition allow identifying information of rather low complexity, i.e., lo-
cating an object or recognizing patterns. Most models, however, go well be-
yond complexity that can be handled by search and recognition. For instance,
a Boolean expression certainly cannot be interpreted just by looking at it and
without deliberate thought. Here, the human brain as “truly generic problem
solver” [13] comes into play. Thereby, cognitive psychology differentiates be-
tween working memory that contains the information is currently processed, as
Qualitative Comparison 7
Prepare
film for edit
Edit footage
offline
Perform
negmatching
Finish on
film
Option 1 Option 2excludes
Option 5 Option 6
Option 7
excludes
excludes excludes
Change options
Options constraints
Context model
shooting_media
delivery_media
Context Variable Range of Values Static/dynamic
static
static
‘TAPE’, ‘FILM’, ‘TAPE ˄ FILM’, ‘NEW_MEDIUM’
AB CDEFG
Adjustment point
Option 2
XOR Prepare tape
for edit
˄
˄
XOR Edit footage
online
˄
˄
CTXT RULE (static):
IF shooting_media = ‘TAPE FILM’
X
Z
C
D
AB
X
Z
A
B
XZ
AXZ B
CTXT RULE (static):
IF delivery_media = ‘TAPE’
Option 5
INSERT
DELETE Finish on
film
EF
Finish on
tape
EF
Option 6
CTXT RULE (static):
IF delivery_media = (‘TAPE’ ‘FILM’)
XOR Finish on
tape
˄
˄
EF
X
Z
E
F
XZ
CTXT RULE (static):
IF delivery_media = ‘NEW_MEDIUM’
Option 7
INSERT
Finish on
new medium
FG
CTXT RULE (static):
IF (shooting_media = ‘FILM’
shooting_media = ‘TAPE FILM‘)
(delivery_media = ‘TAPE’
delivery_media = ‘TAPE FILM‘
delivery_media = ’NEW_MEDIUM’)
Option 3
INSERT Telecine
transfer
DE
CTXT RULE (static):
IF shooting_media = ‘TAPE’
delivery_media = ‘FILM’
Option 4
INSERT Record
digital film
master
DE
CTXT RULE (static):
IF shooting_media = ‘TAPE’
Option 1
INSERT
DELETE Prepare
film for edit
AB
Prepare
tape for
edit
AB
DELETE
Perform
negmatching
CD
INSERT
Edit footage
online
CD
Base process
‘TAPE’, ‘FILM’, ‚TAPE ˄ FILM’
DELETE
Finish on film
EF
Fig. 2. Provop model for the screen postproduction process
well as long-term memory in which information can be stored for a long period
of time [14]. Most severe, and thus of high interest and relevance, are the limita-
tions of the working memory. As reported in [15], working memory cannot hold
more than 7±2 items at the same time. In this context, the concept of mental
effort, i.e., the amount of working memory used, is of interest, as it can be used
to assess understanding. In addition, information held in the working memory
decays after 18–30 seconds if not rehearsed [13]. The importance of the work-
ing memory has been recognized and led to the development and establishment
of Cognitive Load Theory, meanwhile widespread and empirically validated in
numerous studies [36]. The concept of mental effort, i.e., the amount of work-
ing memory used, is of interest, as it can be used to assess understanding. As
discussed in [16], higher mental effort is in general associated with lower under-
standing, or more generally, errors are more likely to occur when the working
memory’s limits are exceeded [17]. Instead of only measuring accuracy or time
for assessing the understandability of a notation, measuring mental effort allows
for a more fine-grained analysis. In particular, when differences with respect to
understandability are small, accuracy may not change, even though the mental
effort changes significantly (cf. [18]). Subsequently, we will discuss three factors
that influence the required mental effort: external memory,abstraction, and the
split-attention effect.
External Memory. First, we would like to introduce a mechanism known
for reducing mental effort, i.e., the amount of working memory slots in use. An
external memory is referred to any information storage outside the human cog-
nitive system, e.g., pencil and paper or a blackboard [17, 13, 19, 12]. Information
taken from the working memory and stored in an external memory is then re-
ferred to as cognitive trace. In the context of a diagram, a cognitive trace would
be, for instance, to mark, update, and highlight information [19]. Likewise, in
8 Victoria Torres et al.
the context of process variants, the model itself may serve as external memory.
For example, when deriving a process variant in C-EPC for a particular context,
the model reader may cross out model elements that have been removed (not re-
quiring their storage anymore in the working memory). Rather, this information
is transferred to the C-EPC model, freeing up working memory capacity.
Abstraction. Basically, the idea of abstraction is to hide information by
aggregation. As irrelevant information can be hidden from the reader, it becomes
easier to focus on relevant information, i.e., abstraction supports the human
mind’s attention management [12], leading to decreased mental effort. Unlike
in C-EPC, where the process family is represented in a single model, Provop
separates the base model from change options which are abstracted via variation
points. This, in turn, simplifies both the base model and the change options,
presumably making both easier to interpret, as attention is not distracted by an
abundance of modeling elements.
Split-Attention Effect. Even though abstraction has been attributed to
reduce mental effort [16, 20, 21], it typically co-occurs with the split-attention
effect, which is known to increase mental effort [22]. In general, the split-attention
effect occurs whenever information from different sources needs to be integrated.
As the human mind can only focus on a single aspect at the same time [23], atten-
tion needs to be constantly switched between the information sources, leading to
increased mental effort. In addition, the task of integrating information is known
to further increase mental effort [22]. To illustrate the split-attention effect, con-
sider the change options in the Provop approach. In this approach, the model
reader has to constantly switch attention between the base model and the change
options in order to extract a new process variant. In addition, the model reader
has to integrate information from change options to the base model, i.e., adding,
removing, moving model elements into the base model, further increasing the
mental effort.
5 Qualitative Comparison
So far we have discussed C-EPC and Provop as representatives of approaches
for modeling BP variability. In the following, we will use concepts from cognitive
psychology to systematically assess differences between these two proposals with
respect to understandability. Since understandability not only depends on the
notation, but also on the type of task to be performed [6, 7], we have structured
the comparison of both approaches along several such tasks. In particular, we
will discuss the task of:
1. extracting a process variant from a configurable process model given a certain
context,
2. understanding the factors driving variability (i.e., the context variables and
their values), and
3. understanding the relationships between process fragments.
Qualitative Comparison 9
In our qualitative comparison, we assume a setting where the model reader
has the models available in paper-based form. We are aware that, even though
our discussion focuses on cognitive aspects only, tool support is indispensable for
working with configurable models. However, to be able to develop effective tool
support, it is essential to know what makes configurable process models hard to
understand. Without a profound discussion, as provided here, tool development
is rather driven by speculation than by systematic consideration.
In the following subsections, we provide a discussion, first for C-EPC and
afterwards for Provop, structured along the following points: First, we will de-
scribe the task in the form of an algorithm describing the steps a model reader has
to perform in order to perform the respective understandability task. The algo-
rithms have been derived in an iterative manner by observing a set of model read-
ers conducting the described tasks. Second, we will illustrate the presented al-
gorithm using the screen postproduction project introduced in Section 2. Third,
we will perform an in-depth analysis to determine the cognitive complexity of
the task.
5.1 Extracting a Process Variant Suitable for a Certain Context
This section discusses first for C-EPC and then for Provop the extraction of a
specific process variant for a given context from the configurable process model
(i.e., domain characteristics of the intended process variant). Such a task is, for
example, important to understand how the process looks like given a certain
context. In addition, it is essential to determine whether a specific business re-
quirement is covered (e.g., when creating or modifying the configurable process
model). Extracting a process variant can also become necessary as part of a more
complex task like understanding the commonalities/differences between a set of
variants. To illustrate the process in both proposals, we assume that a model
reader wants to derive the process variant that relates to the production of a
low-budget project which implies the use of tape as medium for both shooting
and delivery tasks.
5.1.1 Extracting a Process Variant Using C-EPC
To obtain a specific process variant from a configurable process model in C-
EPC, the model reader has to inspect all configurable nodes (i.e., connectors and
functions), evaluate the associated requirements, and configure them according
to the given context. This process is explained in more detail in Algorithm 1.
Example. To obtain the process variant that relates to the low-budget
project in C-EPC (i.e., when shooting and delivery medium is tape) the model
reader starts with the configuration of configurable connector 1. For this pur-
pose, the model reader locates all requirements attached to it and evaluates the
ones of interest to extract the wanted process variant, i.e., reqs 1-3. According
to the given context (i.e., shooting media is tape), configurable connector 1 is
configured as SEQ1aas stated in req. 1. Next, the configuration of configurable
connector 2 has to be performed. In this case, req. 4 determines that the same
10 Victoria Torres et al.
Algorithm 1 Algorithm for extracting a process variant from a C-EPC model
1: for all cn ∈configurable nodes do
2: Locate requirements associated to cn
3: for all r∈associated requirements do
4: Evaluate Boolean expression ex in r
5: if r fits the given context then
6: Configure cn as stated in r
7: Break
8: end if
9: end for
10: if no requirement fits or no requirements are attached then
11: Change cn to a regular (non-configurable) node
12: end if
13: end for
configuration performed to configurable connector 1 should be applied to config-
urable connector 2, i.e., SEQ2ais chosen. Then, the configuration of configurable
connector 3 has to be done. For such purpose, reqs. 5–7 are evaluated. After eval-
uating req. 5 the model reader discovers that SEQ3ashould be chosen. Unlike
reqs. 1–3, reqs. 4–7 are entirely expressed in terms of the structure of the model
(defining implications between alternative process fragments), not providing any
information regarding the context driving the configuration of the variant. This
means that the model reader, in order to understand why this configuration is
performed, has to either remember the decisions previously taken or go back
in the model and revisit the requirements that determined the configuration
of related nodes. Similarly to the configuration performed for configurable con-
nector 2, req. 8 determines that configurable connector 4 should be configured
equally to connector 3, i.e., as SEQ4a. For configuring configurable connector
5req. 9–12 are evaluated. In this case, by evaluating req. 9 the model reader
discovers that it should be configured as SEQ5aand that function finish on tape
should be switched ON. The fact that these requirements include context vari-
ables help the model reader to better understand which configuration should be
taken. Next, the model reader has to decide about the configuration of function
telecine transfer. In this case req. 14 states that this function should be config-
ured as OFF (since connector 3 was configured as SEQ3a). Then, configurable
connector 6 is configured as SEQ6aaccording to req. 15. Finally, function finish
on new medium is switched OFF since function finish on tape has been switched
ON (req. 16).
Cognitive Discussion. Considering the cognitive complexity the use of C-
EPC entails, we can identify three basic operations: locating elements (line 2),
evaluating Boolean expressions (line 4), and adapting the model accordingly
(line 6).
As argued in [24], the more distinct properties a visual element has, e.g.,
shape and color, the easier it is to identify (i.e., to locate the element). In C-
EPC, requirements are represented by white tags, configurable connectors are
represented by white circles with a thick border, whereas configurable functions
Qualitative Comparison 11
are represented by green rounded rectangles with a thick border. Hence, the
reader can draw on two different properties (i.e., color and shape) for identify-
ing distinct modeling constructs presumably requiring a low mental effort. For
identifying whether there are requirements attached to a configurable node (line
2), the model reader can rely on pattern recognition [12] to efficiently perform
this operation (requirements are connected via dotted lines).
The first real challenge occurs when the model reader has to evaluate asso-
ciated Boolean expressions. As they can be arbitrarily complex and have to be
interpreted in the model reader’s mind, presumably a high mental effort can be
expected. While some requirements can be evaluated by just considering their
Boolean expression (i.e., req. 1–3 and 9–12), others depend on decisions previ-
ously taken (i.e., req. 4–8 and 13–16). This requires a bigger cognitive effort by
the model reader, since the model reader has to remember decisions taken for
already configured nodes. For example, when evaluating req. 14 for configuring
function telecine transfer, the model reader has to go back to the related con-
figurable nodes, i.e., to configurable connector 3. In addition, the model reader
has to consider the configuration of functions finish on tape (i.e., req. 9, 10, and
12) and finish on new medium (i.e., req. 12) to understand the semantics of the
configuration. The complexity of evaluating Boolean expressions can differ sig-
nificantly depending on how many related requirements have to be additionally
evaluated and the complexity of them.
Having evaluated the requirements, respective model adaptations have to
be performed (e.g., removing model elements; cf. line 6). Even though these
operations are rather simple the model reader has to keep track of all changes
made during the configuration. Due to the limited nature of working memory
(7±2 slots), it seems essential that the model reader can offload parts of the
working memory to an external memory, e.g., by annotating a print-out of the
model.
5.1.2 Extracting a Process Variant Using Provop
To extract a process variant in Provop, the model reader has to proceed as
follows. First, the model reader has to examine all change options and their
Boolean expressions. Second, the model reader has—based on this evaluation—
to select all change options whose context rule satisfies the given context. Third,
the model reader has to determine for all the selected options, by examining the
constraint model, whether they can be applied considering the already applied
options. Fourth, the model reader has to locate variation points where options
apply, and fifth, the model reader has to mentally integrate the change opera-
tions gathered in the selected change options into the base model. This process
is explained in more detail in Algorithm 2.
Example. To extract a process variant in Provop the model reader first
examines all change options including their associated context rules to determine
which options are applicable for the given context (line 1–6). Based on this, the
model reader then selects options 1 and 5, since only these two satisfy the given
context (lines 3–4) and applies them to the based model (lines 7–14). For this the
12 Victoria Torres et al.
Algorithm 2 Algorithm for extracting a process variant in Provop
1: for all o∈change options do
2: Evaluate Boolean expression ex from associated context rule
3: if ex satisfies the given context then
4: Add o to the selected options list
5: end if
6: end for
7: for all os ∈selected options list do
8: Evaluate constraints between all os
9: if os is not excluded by any of the already applied options then
10: for all ch ∈change operations from os do
11: Locate adjustment points in the base model as specified in ch
12: Apply ch to the base model within the corresponding adjustment points
13: end for
14: end if
15: end for
model reader checks the constraints between the selected options (lines 8–9) and
concludes that both option 1 and option 5 can be applied. The application of
option 1 involves deleting activity prepare film for edit between variation points
A and B, and inserting activity prepare tape for edit instead (lines 11–12). In
addition, activity perform negmatching is replaced by activity edit footage online.
Moreover, the application of option 5 implies the replacement of activity finish
on film by finish on tape between adjustment points E and F (lines 11–12).
Cognitive Discussion. Considering the cognitive complexity of Provop we
can identify two main operations. First, selecting appropriate change options,
and second, applying these change options to the model.
For the identification of relevant change options the model reader inspects
all change options and evaluates whether they are applicable for the current con-
text, i.e., the model reader evaluates the Boolean expression associated with the
change option (lines 1–6 in Algorithm 2). Similar to C-EPC, it can be expected
that the interpretation of such Boolean expressions presumably imposes a high
mental effort. However, unlike C-EPC, in Provop Boolean expressions are always
expressed in terms of context variables. These variables provide semantics to the
change options, helping the model reader to understand the intent of the options.
After having identified relevant change options, the model reader has to apply
them to the base model (cf. lines 7–14). Before applying a change option, it has
to be checked whether the change option is conflicting with previously applied
change options (lines 8–9). This task, however, can easily be accomplished using
Provop’s option constraints, i.e., a set of relationships (e.g., inclusion, exclusion)
between change options targeted to ensure their proper use based on the se-
mantics of the domain. As these option constraints are visually depicted, the
model reader’s recognition capabilities will help to efficiently identify conflicting
options, hence presumably imposing a low mental effort.
Regarding the actual manipulation of the model (lines 11–12), the effort for
integrating the change options into the base model is determined by the change
Qualitative Comparison 13
distance [25] between the base model and the variant to be derived. In other
words, the more modeling elements are added to / removed from the base model,
the more complex the integration task will be. In addition, the type of change
operations contained in the change options influences complexity. For example,
when deleting an activity, respective parts can be removed from the model by
hiding them with a finger on the print-out of the model, or by crossing them
out using a pen. In this way, the model reader does no longer have to keep
this information in his working memory. Rather, the model serves as external
memory, freeing up mental resources. When inserting activities, in turn, this
possibility is not available (except this is done through a supporting modeling
tool) and has to be done in the reader’s mind. Therefore, complexity grows with
the number of changes applied to the base model. Therefore, an optimized design
of the base model that requires minimum changes to derive different variants
presumably requires less effort by the model reader [39]. Altogether it can be
said that most mental effort will presumably be required for evaluating Boolean
expressions as well as conducting model changes.
5.1.3 Discussion
After studying both proposals three major differences can be observed.
First, in a C-EPC model, modeling elements are mainly removed from the
configurable model (with exception of functions when these are configured as
optional, which involve the inclusion of some branching condition in the model).
By contrast, in Provop model elements can either be added, deleted, or moved
during the configuration process. As argued above, the cognitive complexity de-
pends on the type of change operations to be performed (i.e., deletion operations
presumably involve less cognitive effort than insertions or movements).
Second, requirements and configurable nodes are integrated in C-EPC,
whereas change options and the base model are separated in Provop. Similar
to [20, 21], we argue that for small models, C-EPC presumably is easier to un-
derstand, as all the information is integrated and hence in contrast to Provop
no split-attention effect can be expected. However, when model size increases,
models may quickly become too complex resulting in an overload for the model
reader, especially when there are many relationships between alternative model-
ing elements. Here, it can be assumed that the abstraction mechanisms provided
in Provop (i.e., represented by change operations defined separately from the
base model) contributes to retain understandability even for large models.
Third, even though Boolean expressions need to be evaluated in both ap-
proaches, the way they are used by the two proposals differs. In C-EPC, one
of the biggest challenges faced by the model reader relates to the fact that al-
ternatives are usually expressed at the structural level, neglecting the semantics
associated to the different alternatives. This fact involves that, in some cases, the
model reader has to evaluate the Boolean expression at hand, but additionally
has to keep track of previously made decisions. In contrast, in Provop, Boolean
expressions are always expressed in terms of context variables, which contribute
to better understand the semantics of the associated change operations. In ad-
14 Victoria Torres et al.
dition, the concept of options (i.e., grouping of related change operations) as
provided by Provop and the explicit specification of constraints between them
in the constraint model presumably reduces the mental effort required by the
reader for understanding them. Hence, from this point of view, Provop models
presumably impose a lower mental effort on average.
5.2 Identifying Context Variables
This section discusses first for C-EPC and then for Provop how the factors that
lead to variations (i.e., the context variables including their values) can be iden-
tified in a configurable process model. Context variables provide model readers
with the required contextual information to understand the reasons for varia-
tions (e.g., different procedures are followed depending on the shooting media
as well as the delivery media). Context variables also play an important role in
identifying the parts of a configurable process model that need to be changed
(e.g., when a new delivery medium is added). In addition, context information
is also essential when trying to understand differences between variants. More-
over, the different values that a certain context variable can take helps to assess
whether the configurable process model covers all possible alternatives or not
(e.g., whether tape,film, and tape and film are considered as valid shooting me-
dia). To illustrate the process in both proposals we consider the task of extracting
all context variables and all their allowed values from their respective modeling
artefacts.
5.2.1 Identifying Context Variables in C-EPC
To extract context related information from the configurable process model
in C-EPC, the model reader has to locate all the requirements associated with
the configurable elements (i.e., nodes and functions) included in the configurable
process model. Then, based on the Boolean expressions included in these require-
ments, the model reader has to identify the context variables and their allowed
values. This process is explained in more detail in Algorithm 3.2
Example. First of all, the model reader checks requirement 1 and from its
logic predicate extracts context variable shooting media and its first value which
is tape. When examining requirement 2 the model reader realizes that context
variable shooting media has already been used in requirement 1. However, a new
valid value for this variable can be extracted, i.e., film. Similarly, when evaluating
requirement 3 a new value for shooting media is found, i.e., tape and film. On the
contrary, the examination of requirements 4–8 does neither yield any additional
context variable nor any additional value for the already identified variables. By
looking at requirement 9, a new context variable (i.e., delivery media) is discov-
ered and also a first value for it (i.e., tape). The examination of requirements 10
2Note that we assume that context variables are included in the requirements (cf. Sec-
tion 3.1 for details). When using the questionnaire approach together with C-EPC,
in turn, the algorithm looks slightly different, since context variables are modeled
explicitly in the questionnaire and can be extracted from there.
Qualitative Comparison 15
Algorithm 3 Algorithm for extracting process context information in C-EPC
1: for all r∈requirements associated with configurable elements do
2: Identify context variables cv in r
3: for all cv ∈context variables in r do
4: if cv is not yet included in the list of context variables then
5: Add cv to the context variable list
6: end if
7: Identify values for cv
8: for all v∈values for cv do
9: if v is not yet included in the list of allowed values for cv then
10: Add v to the allowed values for cv
11: end if
12: end for
13: end for
14: end for
and 12 shows that three new values for the delivery media variable context are
also valid, i.e., film,tape and film, and new medium. Finally, requirements 13–16
refer to the associated functions and define whether these should be included,
skipped, or optionally included in the model. However, they define neither a new
context variable nor a new value for existing ones.
Cognitive Discussion. From a cognitive point of view, the algorithm can
be analyzed as follows. The identification of requirements, i.e., the identification
of shapes, is performed by recognition (line 1). Similarly, in line 2, the context
variables have to be identified. As illustrated in Fig. 1, context variables have the
following form: context variable =value, e.g., shooting media =’FILM’. Again,
we argue that this identification can mainly be performed by recognition, which
presumably imposes a low mental effort. In lines 4–6, the model reader has to
evaluate whether the variable is already known. Even though the comparison
with existing variables itself can be considered to be rather easy, the 7±2 slots
of working memory available will most likely not be enough to perform this task
for a non-trivial model without the support of some kind of external memory.
Analogously, the extraction of values in lines 7–12 starts with the identification
of values, i.e., recognition. Then, the model reader has to check whether a value
has already been found and add the value to the list of determined values, if
not. Also here we argue that this mental process itself is rather simple, as it
merely includes the comparison of values. However, intermediate results have to
be stored, hence the model reader will most likely need to use external memory
to perform this task.
5.2.2 Identifying Context Variables in Provop
For the extraction of context variables the model reader can directly go to
the context model where all context variables and their allowed values are de-
fined (i.e., table with all context variables + possible values; cf. bottom left part
of Fig. 2). Algorithm 4 describes this process in more detail.
16 Victoria Torres et al.
Algorithm 4 Algorithm for extracting process context information in Provop
1: for all cv ∈context model do
2: Read cv and allowed values
3: end for
Example. In this case just by looking at the context model the model reader
knows that there are just two context variables which are shooting media and
delivery media. While the former can be valued as tape,film, or tape and film
the latter can be valued as tape,film,tape and film, or new media.
Cognitive Discussion. As discussed, in Provop context variables are ex-
plicitly modeled in the context model. This information can directly be extracted
from the model and therefore requires nothing but reading the context values
and allowed values—a low cognitive load can be assumed for this task.
5.2.3 Discussion
As it is explained in this section, for C-EPC models, the extraction of the
context variables and their values is only implicitly available and thus has to
be computed by going through all requirements. On the contrary, in Provop,
this information is explicitly represented in the context model, which facilitates
context variable identification.
5.3 Identifying Relationships between Variable Process Fragments
along a Process Variant
This section deals with the question whether certain relationships (e.g., inclusion
or exclusion) between variable process fragments exist. The identification of these
relationships is very important to ensure that no inconsistencies in respect to the
domain semantics exist in the process model family; these relationships establish
conditions of use of the process fragments when a process variant is derived. To
illustrate how these conditions can be checked we assume that we want to ensure
that telecine transfer is only performed when the shooting media is film or tape
and film and the delivery media is either tape,tape and film or new medium.
5.3.1 Identifying Relationships between Variable Process Fragments
in C-EPC
To accomplish this comprehension task in C-EPC, the model reader has to
locate and select all the requirements that are related (either directly or in-
directly) to the relationship constraint being checked. Then, for each selected
requirement, the model reader has to evaluate the associated Boolean expres-
sion and check whether this satisfy or not the constraint relationship. Algorithm
5 defines this process more in detail.
Example. According to the example, in C-EPC the model reader has to
locate all the requirements that refer to function telecine transfer, and to the
Qualitative Comparison 17
Algorithm 5 Algorithm for identifying relationships between process fragment
alternatives in C-EPC
1: for all r∈requirements do
2: if r has an impact (directly or indirectly) on the condition then
3: Add r to collected requirements list
4: end if
5: end for
6: Analyze relationships between fragments in collected requirements list to decide
whether the relationship constraint is satisfied or not
context variables shooting media (valued as film) and delivery media (valued as
tape or new medium). First of all, the requirement providing the configuration
for function telecine transfer (i.e., requirement 14) states that this function is
only switched on if configurable connector 3 has been configured either as SEQ3b
or as an AND and when functions finish on tape and finish on new medium are
switched on. Regarding the configuration of connector 3, the model reader can
see the configurations to SEQ3bor AND are given by requirements 6 and 7 re-
spectively. In turn, these requirements point to the configuration of configurable
connector 1 and require that is is configured to either SEQ1bor AND. These
configurations are given by requirements 2 and 3. Analyzing these requirements
reveals that respective configurations can be obtained if shooting media is film
(i.e., req. 2) or shooting media is tape and film (i.e., req. 3). At this stage the
model reader already knows that function telecine transfer is only performed if
shooting media is film or tape and film, but not if shooting media is tape. Then,
the model reader has to check whether the requirements also enforce that the
delivery media is either tape or new medium, but not film or tape and film. For
this purpose, the model reader has to check requirements req. 9, 10, and 12. All
three requirements configure configurable connector 5 in such a way that either
SEQ5aor AND are chosen and function telecine transfer can be reached. In
addition, they ensure that functions finish on tape or finish on new medium are
switched ON, which is a requirement for function telecine transfer to be switched
ON. As a result, after checking all the involved requirements, the model reader
can conclude that telecine transfer is only performed when shooting medium is
film or tape and film and delivery medium is tape,tape and film, or new medium.
Cognitive Discussion. This task, as described in Algorithm 5, can be sepa-
rated into two parts. First, the model reader has to locate the set of requirements
involved in the relationship constraint being checked (cf. lines 1–5). In the second
part, the model reader determines whether the selected requirements ensure that
this constraint is well represented. Locating the requirements involves checking
all the requirements attached to the process fragments involved in the relation-
ship constraint being evaluated. In our example, these requirements are 14 (for
function telecine transfer), requirements 6 and 7 (for connector 3), requirements
1 and 3 (for context variable shooting media), and requirements 9, 10 and 12
(for context variable delivery media). While locating requirements constitutes a
relatively simple task, analyzing all the involved requirements (i.e., req. r1,r2,
18 Victoria Torres et al.
. . . rn) to finally find out if the relationship to be checked is well represented
in the model (i.e., to determine if there is a solution to the Boolean formula
r1 ∧r2 ∧. . . ∧rn), presumably requires high mental effort. The complexity
of this task significantly depends on the number of requirements that have to
be evaluated as well as their complexity. In principle, this problem is known to
be np-complete [38] and therefore only possible to be done in one’s mind for
relatively small problems.
Hence, we conclude that for this task the step 1 (identification of require-
ments) may be done manually, but for step 2 (satisfying all requirements), tool
support is indispensable for models that go beyond the complexity of toy exam-
ples.
5.3.2 Identifying Relationships between Variable Process Fragments
in Provop
In Provop relationships between process fragments can be found at two dif-
ferent levels of abstraction. On the one hand, change options allows explicitly
defining inclusion and exclusion relationships by means of delete and insert op-
eration respectively. On the other hand, option constraints allow again defining
relationships, but this time at a higher level of abstraction, i.e., at the options
level. Therefore, the model reader has to analyze both, change options and the
option constraints to find out about relationships between process fragments.
Algorithm 6 defines this process in more detail.
Algorithm 6 Algorithm for identifying relationships between process fragment
alternatives in Provop
1: for all o∈change options do
2: if o refers to the fragments involved in the condition then
3: Add o to collected change options list
4: end if
5: end for
6: Analyze relationships between fragments in change options list to decide whether
the relationship constraint is satisfied or not
Example 1.To test that telecine transfer is only performed when the shoot-
ing media is film or tape and film and the delivery media is either tape,tape
and film or new medium, the the model reader first has to locate the change
options that insert activity telecine transfer (since this activity is not part of the
base model). In the example, change option 5 is the only option inserting activ-
ity telecine transfer. Since there is no further option inserting activity telecine
transfer the question can be answered locally by looking at the context rule as-
sociated with option 5. Evaluating the respective Boolean expression shows that
activity telecine transfer is only inserted in the base model if the shooting media
is film or tape and film and the delivery media is tape,tape and film or new
medium.
Qualitative Comparison 19
In Provop, the complexity for identifying relationships between variable pro-
cess fragments depend on how explicit these relationships are made in the model.
In example 1, the complexity is low since all the information needed can be found
in the change option 5. However, as examples 2 and 3 show, this complexity can
grow when the information is not so explicit in the model.
Example 2. Testing, in turn, whether activity finish on tape and finish on
new medium can co-occur is slightly more complex, since this task cannot be
answered locally by looking at one option only. Like in the previous example we
start by locating all change options where activity finish on tape or finish on new
medium are inserted (i.e., options 5, 6, and 7). Since there is no single option with
both activities we cannot yet decide, but have to analyze the situation further to
determine whether it is possible to execute options 5 and 7 together or options 6
and 7 (in both cases this would result into a co-occurrence of activities finish on
tape and finish on new medium). For this, we first consult the option constraint
model and see that option 5 excludes option 7 and option 6 excludes option 7 as
well. Based on this information we can conclude that the two activities cannot
co-occur. If these relationships would not be defined explicitly in the constraint
model, we would have to check the context rules of options 5, 6, and 7 and check
whether there is a possibility that options 5 and 7 as well as options 6 and 7
co-occur.
Example 3. To answer the question whether a variant exists for which activ-
ity telecine transfer and record digital film master co-occur is even more complex.
This task cannot be answered locally by looking at one option only or using the
constraint model, but requires an analysis of the associated context rules, since
the relationships between these two activities are only implicitly defined. Like
in the previous example we start by locating all change options where activity
telecine transfer or activity record digital film master are inserted (i.e., options
3 and 4). Since the question whether these two activities co-occur cannot be an-
swered by looking at the change options individually, the model reader checks in
the constraint model whether there are any constraints between option 3 and op-
tion 4. Since options 3 and 4 are not mutually exclusive based on the constraint
model, the associated context rules have to be analyzed in detail. This analy-
sis shows that activities telecine transfer and record digital film master cannot
co-occur. Telecine transfer requires shooting media to be film or tape and film,
while record digital film master is only executed if shooting media is tape.
Cognitive Discussion. Similar to the algorithm for C-EPCs, Algorithm 6
can be broken down into two main parts. First, the model reader identifies
relevant change options in lines 1–5. Second, the model reader checks whether
the collected change options satisfy the relationships to be tested. For identifying
all related change options, the model reader has to inspect each change option
individually as no change option can be excluded beforehand. In our example,
all change options that insert activity telecine transfer have to be considered.
Having identified the relevant change options the relationships between them
have to be analyzed (cf. line 6). Depending on the type of question asked, the
decision of whether or not the identified change options satisfy the relationships
20 Victoria Torres et al.
can be made based on a single option (e.g., option 3 in Example 1), by ana-
lyzing the constraint model (e.g., Example 2), or might require analyzing the
context rules associated with the identified change options (e.g., Example 3). If
the question cannot be decided locally based on a single option or based on the
option constraints, like for a C-EPC model, a solution has to be found that sat-
isfies the relationship to be tested. Hence, in general, this task can be considered
np-complete for Provop as well, unless shortcuts are possible.
5.3.3 Discussion
After studying both proposals for identifying relationships between variable
process fragments, one main point can be observed. In both proposals, it is nec-
essary to go through all requirements and change options, respectively, to find
the ones referring to the condition being evaluated. Then, in C-EPC the model
reader has to infer the type of existing relationship between the fragments. In
turn, in Provop, these relationships are potentially explicitly defined as part of
a change option or in the constraint model, which facilitates the analysis. If not,
similarly to C-EPC, the model reader has to evaluate the associated context
rules to find out whether the condition being evaluated is satisfied or not. Put
differently, if the option constraints can be used to determine the relationships
between variable process fragments or the relationships can be answered by a
single change option locally, then the task can be facilitated tremendously. Oth-
erwise it will presumably be, just as for C-EPC, extremely difficult to accomplish
this tasks without tool support.
6 Related Work
Within the business process management community, several proposals have
been developed to deal with the representation of variability in business process
models (e.g., Provop [1], C-EPC [2], PESOA [3], Rule representation and pro-
cessing [4], and PPM [5]). These works take a design-oriented perspective and
provide technical solutions for managing variability; the understandability of the
artifacts created using such approaches is not in the focus. Recently qualitative
evaluations of the C-EPC proposal in form of case studies have been conducted
[26, 27]. However, to the best of our knowledge no work has addressed under-
standability of process model families explicitly. Closely related to this is exist-
ing research on the understandability of process models. For example, in [33],
understandability of process models is approached from a theoretical point of
view. Complementary, several studies report on empirical investigations. Most
approaches thereby employ the concept of metrics computed on structural as-
pects of the process model to assess understandability, e.g., [28, 29, 30, 37, 34].
While metrics seem to be a promising approach to assess model complexity and
understandability, in [6, 7] it is shown that understandability of a process model
significantly depends on the type of question asked. Consequently, a metric will
only be able to roughly estimate understandability. In [31, 20, 21], concepts from
Qualitative Comparison 21
cognitive psychology are used as a tool to discuss understandability of process
models for specific comprehension tasks in the context of both imperative and
declarative process models. In this paper we build upon this work, and extend
it to discuss a selected comprehension task specific to process model families.
Another interesting work to be mentioned is [35], in which empirically validated
guidelines for process modeling, with the aim to improve understandability, are
presented. Unlike in this paper, aspects specifically related to the understanding
of process model families are not addressed.
7 Summary and Outlook
The main goal of this paper is to compare the structural and behavioral ap-
proaches for modeling process model families in terms of understandability. In-
stead of looking at understandability from a broad perspective, the discussion is
centered around the extraction task of a process variant from the modeling arti-
facts produced by C-EPC and Provop. The different approaches are discussed in
terms of different concepts from cognitive psychology based on which the men-
tal effort required to understand the modeling artifacts produced by the two
approaches can be estimated. In turn, this allows us to estimate the understand-
ability of the two approaches for specific comprehension task. The task addressed
in this paper constitutes just a first attempt regarding the investigation of under-
standability of these two approaches. Formal metrics and experiments involving
a large number of subjects and different configurable process models are planned
as future work to empirically test the discussion. Based on the comprehension
tasks presented in this paper, we will conduct a series of experiments to em-
pirically assess the understandability of both approaches and to investigate the
factors that impact understandability of process model families improving the
modeling of such families and facilitating their reuse. For the conduction of such
experiments, we will use the Cheetah Experimental Platform [40] which not only
allows testing the outcome of process modeling (i.e., the created process models),
but also the process of process modeling itself.
References
1. Hallerbach, A., Bauer, T., Reichert, M.: Capturing variability in business process
models: the Provop approach, J. Soft. Maintenance. 22(6–7):519–546 (2010)
2. Rosemann, M., van der Aalst, W.M.P.: A configurable reference modeling language.
Inf. Systems 32(1):1–23 (2007)
3. Puhlmann, F., Schnieders, A., Weiland, J., Weske, M.: Variability mechanisms for
process models. Technical report, BMBF–Project (2006).
4. Kumar, A., Wen, Y.: Design and management of exible process variants using tem-
plates and rules. Int. J. Comput. Ind. 63(2), pp. 112–130 (2012).
5. Pascalau, E., Awad, A., Sakr, S., Weske, M.: On Maintaining Consistency of Process
Model Variants. BPM Workshops, 289-300 (2010).
22 Victoria Torres et al.
6. Figl, K., Laue, R.: Cognitive Complexity in Business Process Modeling. In Proc.
CAiSE’11, 452–466.
7. Melcher, J., Detlef, S.: Towards Validating Prediction Systems for Process Under-
standability: Measuring Process Understandability. In Proc. SYNASC’08, 564–571.
8. La Rosa, M., Lux, J., Seidel, S., Dumas, M., ter Hofstede, A.H.M.: Questionnaire-
driven Configuration of Reference Process Models. In Proc. CAiSE’07, 424-438.
9. van der Aalst, W.M.P., Dreiling, A., Gottschalk, F., Rosemann, M., Jansen-Vullers,
M.H.: Configurable process models as a basis for reference modeling. In BPM Work-
shops, LNCS vol. 3812, 512–518 (2005).
10. Vervuurt, M.: Modeling business process variability: a search for innovative solu-
tions to business process variability modeling problems”. Student Theses of Univer-
sity of Twente. October 2007.
11. Hallerbach, H., Bauer, T., Reichert, M.: Issues in Modeling Process Variants with
Provop. Business Process Management Workshops, 2008, pp. 56–67 (2008)
12. Larkin, J.H., Simon, H.A.: Why a Diagram is (Sometimes) Worth Ten Thousand
Words. Cognitive Science, 11(1):65–100 (1987).
13. Tracz, W. J.: Computer programming and the human thought process. Software:
Practice and Experience, 9(2):127–137 (1979).
14. Paas, F., Tuovinen, J.E., Tabbers, H., Van Gerven, P.W.M.: Cognitive Load Mea-
surement as a Means to Advance Cognitive Load Theory. Educational Psychologist,
38(1):63–71 (2003).
15. Miller, G.: The Magical Number Seven, Plus or Minus Two: Some Limits on Our
Capacity for Processing Information. The Psychological Review, 63(2):81–97 (1956).
16. Moody, D. L.: Cognitive Load Effects on End User Understanding of Conceptual
Models: An Experimental Analysis. In Proc. ADBIS’04, 129–143.
17. Sweller, J.: Cognitive load during problem solving: Effects on learning. Cognitive
Science, 12(2):257–285 (1988).
18. Zugal, S., Pinggera, J., Weber, B.: The Impact of Testcases on the Maintainability
of Declarative Process Models. In Proc. BPMDS’11, 163–177.
19. Scaife, M., Rogers, Y.: External cognition: how do graphical representations work?
Int. J. Human-Computer Studies, 45(2):185–213 (1996).
20. Zugal, S., Pinggera, J., Mendling, J., Reijers, H.A., Weber, B.: Assessing the Im-
pact of Hierarchy on Model Understandability-A Cognitive Perspective. In Proc.
EESSMod’11, 123–133.
21. Zugal, S., Soffer, P., Pinggera, J., Weber, B.: Expressiveness and Understandabil-
ity Considerations of Hierarchy in Declarative Business Process Models. In Proc.
BPMDS’12, 167–181.
22. Sweller, J., Chandler, P.: Why Some Material Is Difficult to Learn. Cognition and
Instruction, 12(3):185–233 (1994).
23. Feldmann Barrett, L., Tugade, M. M., Engle, R. W.: Individual Differences in
Working Memory Capacity and Dual-Process Theories of the Mind, Psychol Bull.
130(4):553-573 (2004).
24. Moody, D.L.: The “Physics” of Notations: Toward a Scientific Basis for Construct-
ing Visual Notations in Software Engineering. IEEE Trans. Soft. Eng. 35(6):756–779
(2009).
25. Li, C., Reichert, M., Wombacher, A.: On Measuring Process Model Similarity
Based on High-Level Change Operations. In Proc. ER’08, 248–264.
26. L¨onn, C.M., Uppstr¨om, E., Wohed, P., Juell-Skielse, G.: Configurable Process Mod-
els for the Swedish Public Sector. In Proc. CAiSE’12, 190–205.
Qualitative Comparison 23
27. Gottschalk, F., Wagemakers, T., Jansen-Vullers, M., van der Aalst, W., La Rosa,
M., van Eck, P., Gordijn, J., Wieringa, R.: Configurable Process Models: Experiences
from a Municipality Case Study. In Proc. CAiSE’09, 486–500.
28. Mendling, J., Reijers, H.S., Cardoso. J.:What Makes Process Models Understand-
able? In Proc. BPM’07, 48–63.
29. Vanderfeesten, I., Reijers, H.A., van der Aalst, W.M.P.: Evaluating workflow pro-
cess designs using cohesion and coupling metrics. Int. J. Comput. Ind. 59(5):420–437
(2008).
30. Reijers, H.A., Mendling, J.: A Study into the Factors that Influence the Under-
standability of Business Process Models. SMCA 41(3):449–462 (2011).
31. Zugal, S., Pinggera, J., Weber, B.: Assessing Process Models with Cognitive Psy-
chology. In Proc. EMISA’11, 177–182.
32. La Rosa, M., van der Aalst, W.M.P., Dumas, M., ter Hofstede, A.H.M.:
Questionnaire-based variability modeling for system conguration. Software and Sys-
tem Modeling 8(2), pp. 251–274 (2009).
33. Becker, J., Rosemann, M., Uthmann, C.: Guidelines of Business Process Modeling.
In Proc. BPM’00, 30–49 (2000).
34. Melcher, J., Mendling, J., Reijers, H.A., Seese, D.: On Measuring the Understand-
ability of Process Models. In Proc. BPM Workshops, 465–476 (2009).
35. Mendling, J., Reijers, H.A., van der Aalst, W.M.P.: Seven process modeling guide-
lines (7PMG). Information & Software Technology, 52(2):127-136 (2010).
36. Bannert, M.: Managing cognitive load–recent trends in cognitive load theory.
Learning and Instruction, 12(1):139–146 (2002).
37. Cardoso, J.: Process control-flow complexity metric: An empirical validation. In
Proc. IEEE SCC’06, 167–173 (2006).
38. Cook, S.: The complexity of theorem proving procedures. In Proc. STOC’71. 151-
158 (1971).
39. Li, C., Reichert, M., Wombacher, A.: Mining business process variants: Challenges,
scenarios, algorithms. Data Knowl. Eng. 70(5): 409–434 (2011).
40. Pinggera, J., Zugal, S., Weber, B.: Investigating the process of process modeling
with Cheetah Experimental Platform. In Proc. ER-POIS’10, 13–18 (2010).
