Towards Enhanced Guiding Mechanisms in VR Training Through
Process Mining
Enes Yigitbas, Sebastian Krois, Sebastian Gottschalk and Gregor Engels
Institute of Computer Science, Paderborn University, Zukuntsmeile 2, Paderborn, Germany
Keywords:
Virtual Reality, Process Mining, Usability Evaluation.
Abstract:
Virtual Reality (VR) provides the capability to train individuals to deal with new, complex, or dangerous
situations by immersing them in a virtual environment and enabling them to learn by doing. In this virtual
environment, the users usually train a sequence of different tasks. With that, most VR trainings have an
underlying process that is given implicitly or explicitly. Although some training approaches provide basic
guidance features, when analyzing the execution of the training, the process itself is often not considered,
even if the process is one of the primary aspects to train in many cases. In this paper, we present VR-ProM,
a framework that enables to use process mining techniques by supporting logging, analysis of execution logs
of training sessions, and provision of guiding mechanisms to enhance VR training applications. To evaluate
our framework and to investigate whether the integration of process mining techniques enables us to support
the enhancement of VR-based training applications, we performed a two-staged user study based on a VR
warehouse management training application. To analyze the effectiveness and subjective usability of the VR
training, we performed two rounds of user studies and compared the results before and after we integrated the
guiding mechanisms driven by process mining. Initial usability evaluation results show that with the help of
VR-ProM the trainees made 40% fewer mistakes in the example VR training application and that the overall
user satisfaction could be increased.
1 INTRODUCTION
Recent advances in Virtual Reality (VR) technology
and the increased availability of VR-equipped de-
vices enable a wide range of applications in various
domains such as medicine (Gurusamy et al., 2009),
robotics (Yigitbas et al., 2021), or marketing (Alca
˜
niz
et al., 2019). This work focuses on the use of VR for
training which is defined as the provision of knowl-
edge and skills in an interactive manner (Antona-
copoulou, 2001). In this context, VR provides the
capability to train individuals to deal with new, com-
plex, or dangerous situations by immersing them in
a virtual environment and enabling them to learn by
doing.
While VR-based trainings usually rely on a pro-
cess model which is given implicitly or explicitly for
the task to learn, none of the existing VR-based train-
ing approaches make use of this process knowledge
to improve the guidance of the training application
by applying process mining techniques. As a con-
sequence of this, the users of such VR-based train-
ings have to face a ”one-size-fits-all” training with
fixed guidance or a freestyle training approach with
no guidance. However, due to learning style and ex-
perience level differences as well as task needs, it is
important to consider the underlying processes and
analyze their execution logs for engineering effective
and user-friendly VR-based training applications. For
this purpose, Process Mining can be seen as a promis-
ing technique to enhance the guiding mechanisms in
VR-based trainings. However, the integration of pro-
cess mining techniques and their implication for VR-
based training applications have not been broadly re-
searched so far. Although process mining has a big
potential to analyze and optimize training processes,
a systematic method how to integrate process mining
techniques for VR-based trainings is missing. In ad-
dition to that, it is important to investigate whether
process mining techniques are beneficial to enhance
the guiding mechanisms in VR-based trainings. To
the best of our knowledge, there is no approach ex-
isting that deals with the application of process min-
ing to enhance VR-based training applications. This
together leads to the research question of our work:
”How can we use process mining techniques to en-
152
Yigitbas, E., Krois, S., Gottschalk, S. and Engels, G.
Towards Enhanced Guiding Mechanisms in VR Training Through Process Mining.
DOI: 10.5220/0011651600003417
In Proceedings of the 18th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2023) - Volume 2: HUCAPP, pages
152-159
ISBN: 978-989-758-634-7; ISSN: 2184-4321
Copyright
c
2023 by SCITEPRESS Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)
hance the guiding mechanisms in VR-based train-
ings?”
To answer this question, our work contains the
following contributions. Firstly, we have developed
a conceptual solution for the integration of process
mining techniques in VR-based training applications.
Secondly, we have implemented the conceptual so-
lution as a framework, called VR-ProM, for Unity
1
projects to show its functionality in action. To eval-
uate our process mining solution approach, we have
used an example VR-based training application from
the domain of warehouse management and conducted
a two-staged user study.
The rest of the paper is structured as follows. In
Section 2, we present and discuss the related work.
In Section 3, we describe the conceptual solution and
implementation of our process mining approach for
VR-based training applications. In Section 4, we
present the user study and discuss the main results of
the usability evaluation. In Section 5, we conclude the
paper and give an outlook for future work.
2 RELATED WORK
Virtual Reality-based training applications have been
discussed in the past for various application domains.
In the following, we briefly draw on prior research
into Virtual Reality Trainings, Execution and Interac-
tion Logging as well as Process Mining for Training
Scenarios.
2.1 Virtual Reality Trainings
Virtual Reality trainings have been used for several
years now. For example, Liang et al. (Liang et al.,
2019) developed a training game for miners, which
teaches them how to spot loose rocks in underground
mines to decrease rock-related hazards. Another ex-
ample is given by Shen et al. (Shen et al., 2019) who
simulated a marine ship to train marine engineers to
work on a real ship. Furthermore, Wang et al. (Wang
et al., 2017) created a surgeon training for medical
students. Based on this approach, they can train surg-
eries in virtual reality before they start operating on a
living patient.
While the above-mentioned VR-based training ap-
plications support the training process for a specific
domain-relevant task, they do not make use of data
logging or even if they record data, the results are not
used to improve the training application by adjusting
the guiding mechanisms.
1
www.unity.com
2.2 Execution and Interaction Logging
In (Vidani and Chittaro, 2009), Vidani et al. present
an approach where the combined use of task models
and logs of user interactions are investigated to ana-
lyze the training process of serious games for emer-
gency medical procedures.
Furthermore, Harms (Harms, 2019) presents an
automated usability evaluation approach for VR ap-
plications. To enable logging, he presents a tool that
crawls through the objects placed in a scene of the VR
application to evaluate and automatically detect rele-
vant log actions performed by the user.
A further approach where monitoring of execu-
tion logs of VR applications is considered is presented
by Zahabi et al. (Zahabi and Abdul Razak, 2020).
Here, the authors argue that VR trainings should be
adapted based on the user’s capabilities, performance,
and needs. For this purpose, they provide a system-
atic literature review and a framework for adaptive
VR-based training including performance measures,
adaptive logic, and adaptive variables.
While the above-mentioned approaches enable
conformance checking by taking a process or inter-
action log and comparing it to an existing process or
task model, they do not support process discovery and
enhancement which is possible using process mining.
2.3 Process Mining for Training
Scenarios
Dolak et al. (Dolak, 2019) took the logs of the on-
line education platform MOODLE
2
to analyze the
students’ behavior following an online course.
In the context of warehouse management,
Paszkiewicz and Zbigniew (Paszkiewicz, 2013) ana-
lyzed the product management process of a mattress-
producing company. The process describes how to
treat pallets with freshly produced mattresses until
they are shipped to the customer.
Fern
´
andez-Gallego et al. (Fern
´
andez-Gallego
et al., 2013) present a learning analytics framework
for 3D educational virtual worlds that focus on dis-
covering learning flows and checking their confor-
mance through process mining techniques. The core
of this framework is an Opensim-based virtual world
platform that has the ability of monitoring and regis-
tering the events generated by students and teachers.
Furthermore, Cerezo et al. (Cerezo et al.,
2020) introduce a process mining approach for self-
regulated learning assessment in the context of e-
learning.
2
https://moodle.com/
Towards Enhanced Guiding Mechanisms in VR Training Through Process Mining
153
One approach where VR and process mining are
combined is presented by Rold
´
an et al. (Rold
´
an et al.,
2019). In this approach, the authors describe a train-
ing system for Industry 4.0 operators in complex as-
semblies based on VR and process mining. While the
main idea of this approach is similar to ours, they are
not focusing on the improvement of the VR interface
by adjusting the guiding mechanisms to the needs of
the end-users.
3 CONCEPT AND
IMPLEMENTATION
To answer our research question and to integrate pro-
cess mining techniques for the analysis of VR-based
trainings, we have developed a VR process mining
analysis framework called VR-ProM. A high-level ar-
chitectural overview of VR-ProM is shown in Figure
1.
VR Application
Task Objects
Gateway Objects
Guiding
Component
Logger
Component
Performs
action
Reports
action
Check
condition
status
Activate
help
Report action
to log
VR-Prom Framework
Process
Model
Log
Process Mining
Analysis
Produces
Updates
Relies on
Relies on
Used for
analysis
Creates/Updates
VR End-User
Figure 1: Architectural Overview.
In VR-ProM, we assume that there is already an
existing VR-based training denoted as VR Application
which relies on an explicit Process Model. The VR
End-User performs an action in the VR-based train-
ing that is forwarded to the Guiding Component and
logged by the Logger Component. In addition to this,
the action is evaluated. Therefore, the Guiding Com-
ponent needs to know where in the process the user
is. To determine that, every application needs cus-
tom Check condition status operations which deter-
mine which path the user has to follow. When the user
executed the right task, the Guiding Component also
needs to check if the task was executed completely. If
one of these conditions does not hold, the help mecha-
nism can be activated. For every application, the help
mechanisms need to be implemented individually. It
also needs to be defined, when the help is to be ac-
tivated (i.e. which preconditions need to hold before
showing the help). That can be based on Process Min-
ing Analysis, e.g. by determining how many mistakes
may happen before the help is activated.
The Logger Component allows developers to cap-
ture the users’ behavior in a format that can be used
directly for Process Mining Analysis. The Guiding
Component takes a process model as input. Then, it
supervises if the user acts according to the specified
process model. Additionally, the opportunity to eas-
ily implement guiding mechanisms into a VR Applica-
tion is possible. Which guiding mechanisms are used
and at what point they are applied can be based on
the results of a Process Mining Analysis. The user’s
behavior can be continuously logged so that we can it-
eratively use process mining to successively improve
the guiding. In the following sections, each compo-
nent of the architectural overview will be described in
more detail.
3.1 VR Application - Warehouse
Management Training
For an illustration of the main concepts of the VR-
ProM framework, we firstly describe the VR applica-
tion that serves as a basis and motivational real-world
example for our solution idea. In our case, we take
an existing VR application from the domain of ware-
house management training. In the area of logistics,
warehouse management involves a complex process
that consists of different order picking tasks. Due to
this complex process, stock discrepancies and mis-
placed wares are typical problems that often occur.
To overcome this problem, we have developed a VR
training application that integrates an existing ware-
house management system and trains typical order
picking processes. In the following, the main func-
tionality and structure of this VR-based training ap-
plication will be briefly described to guarantee a bet-
ter understanding of the forthcoming concepts con-
cerning the integration of process mining. First of all,
Figure 2 shows how the warehouse is structured.
The core objects in a warehouse are the items
stored in the warehouse. There are different types of
items. To limit the possible item types, we decided
there should be Small Items and Big Items. With that,
we have two item categories that need to be treated
differently. The small items can be grabbed and car-
ried by hand. We also use a Carrying Robot in which
the user can put small items, and which can follow
them through the warehouse. The robot allows each
user to collect multiple small items at the same time,
so they do not have to walk the same way again for ev-
ery item. Small items can be fragile or robust. These
attributes do not change their behavior but need to be
considered when putting them into the Picking Cart
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
154
3
5
2
4
7
8
6
1
6 6 6
66
66
3
3
3
3
3
3
3
3
3
3
3
3
3
8
8
888
3
2
1
6
8
7
4
5
VR-End User
Order Display
Small Items
Carrying Robot
Picking Cart
Big Items (Pallets)
Forklift
Pallet Places
Figure 2: Floor Plan of the Warehouse Building.
for shipping. The picking carts are boxes, in which
the user puts items ready to be shipped. To keep it
neat, there is one picking cart for every order. A pick-
ing cart itself is divided into two sections, one for
fragile items and one for robust ones. When placing
items in the picking cart, the user has to place them
according to their attributes. Big items are too heavy
to carry by hand. They are placed on a pallet which
requires the player to use a Forklift for moving them.
To increase the immersion, the Forklift is controlled
by the user’s movement, not by pushing buttons on
the controller. Therefore, it has a handle that can be
grabbed. By moving the handle, the user can move
the Forklift. The Forklift additionally needs two but-
tons to enable the up and down lifting of its forks to
carry the pallets. For shipping, the pallets need to be
placed in designated Pallet Places.
3.2 VR-ProM Framework
In the following, we describe our novel VR-ProM
framework
3
in more detail by focusing on its two
main components Logger and Guiding.
3.2.1 Logger Component
For analyzing the user’s behavior and training per-
formance we can log the events occurring during the
training. The Logger Component produces a log that
conforms to the XES standard
4
, so we can use it with
most of the existing process mining tools. To be able
to easily combine traces from multiple computers into
one log, the Logger Component writes down every
playthrough in its own file and combines them to the
actual log before ending the application. With that,
we can use multiple traces and combine them into a
3
https://github.com/Quegg/VR-Prom
4
https://xes-standard.org
BPMN
Generate class stubs
h
Figure 3: Component Overview.
single log later. To be able to use the Logger Compo-
nent outside this application example, it is indepen-
dent of the application. First, we need the Logger
Component itself, which can be called to perform the
logging. To ensure the last requirement, only one in-
stance of the logger may be running at the same time.
It needs at least three functions. One for initializing
the logger and opening a new trace, one to close the
log, and finally one for logging itself. The last one
gets the event and logs it accordingly. Every event
needs its own class where the class attributes are the
attributes occurring in the log. Therefore, all events
are an XesEvent. Every public field which imple-
ments the XesAttribute interface in the event class will
be logged as an XesAttribute or XesExtension (when
possible). We support primitive types and collections
such as lists to also allow nested attributes.
3.2.2 Guiding Component at Design-Time
Besides the logging functionality described above,
our VR-ProM framework provides a Guiding Com-
ponent that enables assistance mechanisms in a VR-
based training application to better help and assist the
trainees. It is realized with the GuidingController
script in the framework. It consists of two groups
of functionalities. The first group includes features
needed in the Unity editor while developing the VR-
based training application. The second handles the
guiding at runtime. Before integrating help mecha-
nisms, we need to know the process we want to teach.
Many tools allow the specification of process mod-
els. In the current version of the VR-ProM frame-
work, we support process models saved as .bpmn file
(e.g. created by the Camunda Modeler
5
). The process
may contain tasks, exclusive gateways, start, and end
events. To load the process model from the .bpmn
file, we need a BPMN Parser which is used in both
functionality groups.
5
https://camunda.com/download/modeler/
Towards Enhanced Guiding Mechanisms in VR Training Through Process Mining
155
Figure 3 shows the top-level components of the
Guiding Component and the very basics of their com-
munication. The Guiding Controller uses the BPMN
Parser to get the process as input. Then, it creates
an empty stub file for the following elements of the
reference training process.
Task: For every task, we need an event to log and
an object which is loaded in the VR Application to
actually interact with the trainee and all objects in
the warehouse. This object will later contain the
guiding mechanisms for its task.
Gateway: Since a gateway is not an event that can
be logged, we do not need a logging event. But we
do need an object which is loaded in the scene,
just like the task does. This is necessary, as the
gateways have to check the state of their condi-
tions to determine the next task to execute.
Error Events: Error events are not parsed from the
process. They can be added manually afterward.
When adding an error event, we create a logging
event and a class to be loaded in the VR Applica-
tion as well.
After that, the developers can adjust the stubs, so they
fit into their application. The core class of this project
is the GuidingController. It handles the generation of
all class stubs and manages the whole guiding during
the simulation. To activate the GuidingController, the
corresponding script needs to be added to a GameOb-
ject. In Unity, a GameObject is the most basic compo-
nent which can be seen as a container holding every-
thing being present in a scene ranging from visible 3D
models to invisible scripts performing some action.
3.2.3 Guiding Component at Run-Time
In addition to the components we described above, we
implemented a very lightweight State Machine (see
Figure 3). It holds the information about the whole
process and knows the task the user is currently work-
ing on. It does not need to execute anything but it
has to know the current state and provide some util-
ity functions on the process, e.g. getting all possible
tasks following the current one.
Here, we first check the simplest case of the next
element being a task. In that case, we do not need
to check any conditions and simply return to the next
task. If the next element is a gateway, we check its
conditions implemented by the developers. A gate-
way provides a method to CheckConditions. It re-
ceives a list of all possible next elements and returns
the name of the correct one. To use that, we first re-
trieve the names of all the following elements from
the State Machine. Then, we pass this list to the gate-
way and use the returned object. If it is a task, we
return this task. Otherwise, we check the conditions
of the next gateway the same way until we eventually
reach a task.
User management is optional and can be used to
assign the playthrough results to user IDs to group or
compare them later.
To show help to the user, the developers can im-
plement custom help features (navigation arrows, ob-
ject highlighting, etc.) for every task and error. If
no custom help is provided, we provide generic help
as a fallback. This generic help will be based on the
given process and shows a simplified part of the pro-
cess model depicting where the user currently is, what
are the possible next tasks, and which of the next tasks
is the right one. Figure 4 shows what this looks like
in the simulation.
Figure 4: Exemplary Tasks to Execute During the Training
Process.
To generate this view, we take the last executed
task as input and display all possible next tasks to-
gether with the paths leading to them. We show the
exclusive gateways as black circles which divide the
path. If a path has a condition, it is shown using a
speech bubble. To help the user orientate in this view,
we added a red outline on the current task to execute.
3.3 Process Mining Analysis
With all the concepts described above, we almost have
all parts together to integrate process mining results
into VR trainings. The only thing missing is the pro-
cess mining analysis itself. Using process mining, we
have three different techniques at hand. The first tech-
nique, we can use, is process discovery, to see how the
process is executed in real life. This is a generaliza-
tion of all recorded paths in this log. So we can see,
how a general user executes the process. With that,
we can find out where problems occur, or the correct
order is not clear to the user. The second technique
is conformance checking. As its purpose is to show
where the execution in the log deviates from the de-
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
156
sired log, this seems to be suitable for our approach.
As, in our case, the process executions strongly de-
viate from each other and the given process, we can
see 0% accordance of the log to the process. These
results do not help to detect problems, so we will not
use conformance checking in our analysis. The third
technique, enhancement, is used to enhance the dis-
covered process map, so we can also consider the tim-
ing and frequency of event occurrences. To perform
the process mining analysis, we use the academic ver-
sion of Disco (G
¨
unther and Rozinat, 2012). The de-
rived results of the process mining analysis will be de-
scribed in the next section which deals with the eval-
uation of our VR-ProM framework.
4 EVALUATION
To evaluate our framework VR-ProM and to investi-
gate whether the integration of process mining tech-
niques supports the enhancement of VR-based train-
ing applications, we performed an initial user study
based on the described warehouse management train-
ing application. In the following, first, we describe the
evaluation setup of our user study. Then, we describe
the guiding mechanisms that were determined based
on the application of process mining. Those guiding
mechanisms were then integrated into the VR-based
training to increase its efficiency, effectiveness, and
user satisfaction. To check this, a second user study
took place.
4.1 Evaluation Setup
To evaluate the benefit of VR-ProM, we have con-
ducted a two-staged user study based on the VR ware-
house management training application. In the first
stage, the users performed the training where no guid-
ing mechanisms were integrated. They needed to
figure everything out on their own. The application
logged the user’s behavior, and we used these event
logs to determine the values for the KPIs. Those KPIs
were Efficiency (how long needs a user to complete
the training), Error Rate (how often does the user do
anything wrong), and the Subjective Usability (what
is the SUS score the user gave for the training). Then,
we analyzed the generated process map to identify
problems and determine guiding mechanisms to avoid
them. As soon as the integration was finished, a sec-
ond group of users performed the training, but this
time there were guiding mechanisms available. Due
to the COVID-19 pandemic, we decided to perform
the evaluation in a remote setting.
Due to the short-term conversion into remote eval-
uations, five participants completed the training in the
first evaluation. They were aged between 18-46 years,
two were female and three were male. They had dif-
ferent levels of technical knowledge concerning VR
technology. In the second evaluation, 12 participants
completed the training. Four were female and eight
were male. They were aged between 16-56 years and
had also different levels of technical knowledge about
VR technology.
4.2 Guiding Mechanisms
To identify bottlenecks in the training process and
thus determine possible guiding mechanisms, we per-
formed a frequency analysis in Disco. In the follow-
ing, exemplary remarkable cases are listed and will be
shortly discussed.
Some items were removed from the picking cart.
One reason for that is that the users placed the
wrong items into the picking cart and had to re-
vert this for completing the order. Another reason
could be that the users had to grab the item to scan
it as they forgot it before.
There were more items placed in the robot than re-
moved. All the users completed the training with
correct results (i.e. they placed all needed items
in the picking carts). This means they collected
some items they did not need at all. Reasons for
this could be that the users did not know which
item to collect or how many of them, so they took
it anyway.
The users let some items fall on the ground. This
can have many reasons. For example, they did
not fully understand the controls so an item falls
down, or they picked the wrong item.
The forklift’s barcode was rarely scanned by the
users. So the users either did not realize they need
to scan it, or they forgot it.
Often, the forklift hit an object. It is expected,
that users with no or little experience in VR con-
trols/forklift driving will hit some objects. But in
five playthroughs, the users hit objects 220 times.
Most likely, the users did not understand well
enough how to control the forklift properly.
Concluding the insights gathered above, there are
the following problems. First of all, it is not clear
which items the user has to collect. Furthermore, the
barcode scanner is not used properly or many bar-
codes are not scanned. In addition to that, when plac-
ing items in the picking cart, it is hard to distinguish
between fragile and robust items. Finally, it could be
observed that the forklift’s main controls are not clear
to the users.
Towards Enhanced Guiding Mechanisms in VR Training Through Process Mining
157
The most obvious problem is, that the tasks are not
executed in the correct order. Considering these prob-
lems, we came up with the following guiding mecha-
nisms.
Outline the items to pick up/the objects to interact
with
Outline the barcode scanner and show the buttons
to activate it
Outline the barcodes to scan
Show the way the user has to go/drive
Show the forklift’s controls
Show where and how to place the small items
(fragile or robust) in the picking cart
When showing help, the framework will additionally
show where in the process the user is, what tasks pos-
sibly follow the last one, and which one of them is the
right one to execute.
4.3 Result Comparison
As our goal was to analyze whether our framework
VR-ProM indeed enhanced the VR training applica-
tion, we had to check if the users achieved better re-
sults in the second evaluation. To do so, we first need
to clarify what better results mean. Regarding Effi-
ciency and the Error Rate, we consider lower values
as better values. These two KPIs can be summarized
as the users’ Performance. So a better Efficiency and
Error Rate result in a better Performance. For the
Subjective Usability, we use the SUS scale. Hence,
a higher value corresponds to a better Subjective Us-
ability.
First, we compare the duration shown in Figure
5. We see that in the second evaluation the users
took 10% longer to complete the training, on aver-
age. However, we see that the second results vary
in a broader range, as some users can take more ad-
vantage of the guiding mechanisms than others. The
single dots in Figure 5 denote single values which are
far apart from the others. All in all, this means, that
the Efficiency was slightly worse in the second evalu-
ation.
The second value to consider is Subjective Usabil-
ity. For this, every user filled out the SUS question-
naire. With that, we determine the values shown in
Figure 6. Here, we see that the maximum rating is
equal. But there are fewer low ratings and the ratings
overall vary less. The average SUS score increased
by 6%. This growth itself is not very expressive. But
with also comparing the distribution, we can conclude
the Subjective Usability slightly increased.
Figure 5: Duration. Figure 6: SUS Score.
Figure 7: Error Rate Full. Figure 8: Error Rate.
Now, we compare the Error Rate. For our evalua-
tion, there were two different error rates. Basically, an
error occurs when the user did something that was not
meant to do. This can be an error event (something
that should never happen) as well as a normal task ex-
ecution but of a task which should not be executed at
the current time. Error Rate Full contains all these
events. We excluded the ForkLiftHitsObject Event for
our main Error Rate. The forklift’s movement was
very hard to control for the players. They often expe-
rienced motion sickness and could not concentrate on
steering properly. Additionally, in a real-world ware-
house, the employees are familiar with controlling the
forklift. We do not want to train driving a forklift but
help the user learn the process. For completeness,
we provide both datasets here. We see that in both
datasets, the minimum, maximum, and average error
rate is much smaller in the second evaluation. For ex-
ample, the average amount of errors for the main Er-
ror Rate decreased by 40%. However, the results of
the second evaluation have a higher standard devia-
tion in both datasets. So we can say, that we achieved
a better Error Rate in the second evaluation.
HUCAPP 2023 - 7th International Conference on Human Computer Interaction Theory and Applications
158
4.4 Discussion and Threats to Validity
To summarize the main results of our usability eval-
uation, Regarding Efficiency, we found out, that the
users needed on average 10% longer to complete the
training with the integrated help. However, consid-
ering the Error Rate, we can see improvements when
providing help during the training. The Error Rate de-
creased by 40% in the second playthrough compared
with the first. The Usability, measured with the SUS
scale, improved by 6%. This is only a minor change,
but we can see that, in general, there were fewer low
scores.
However, an important threat is the limited num-
ber of participants. Here, we need larger experiments
with more heterogeneous groups to derive statistically
representative results that help us to generalize our
ideas for various application domains.
5 CONCLUSION AND OUTLOOK
In this paper, we have introduced the VR-ProM
framework that supports the logging of VR-based
training applications and produces log data in stan-
dardized XES format that can be analyzed based on
existing process mining tools. Furthermore, VR-
ProM provides generic and flexible guiding mecha-
nisms to improve the help and guiding mechanisms
in VR-based training applications based on the pro-
cess mining results. Based on an initial evaluation we
have shown the benefit of our VR-ProM framework
by applying it to a VR warehouse management train-
ing application.
While this shows the potential of process mining
for VR-based training applications, further steps are
needed to establish the application of process min-
ing techniques for VR technology. First of all, larger
evaluations with more participants are required. Fur-
thermore, to see the full potential of our solution ide,
the help mechanisms need to be iteratively adjusted
and evaluated. Finally, it would be beneficial to have
a process mining solution that works out of the box
and supports an automated integration in various VR-
based training applications.
REFERENCES
Alca
˜
niz, M., Bign
´
e, E., and Guixeres, J. (2019). Virtual re-
ality in marketing: a framework, review, and research
agenda. Frontiers in psychology, page 1530.
Antonacopoulou, E. P. (2001). The paradoxical nature of
the relationship between training and learning. Jour-
nal of Management Studies, 38(3):327–350.
Cerezo, R., Bogar
´
ın, A., Esteban, M., and Romero, C.
(2020). Process mining for self-regulated learning
assessment in e-learning. Journal of Computing in
Higher Education, 32(1):74–88.
Dolak, R. (2019). Using process mining techniques to dis-
cover student’s activities, navigation paths, and behav-
ior in LMS moodle. In Rønningsbakk, L., Wu, T.,
Sandnes, F. E., and Huang, Y., editors, ICITL, volume
11937 of LNCS, pages 129–138. Springer.
Fern
´
andez-Gallego, B., Lama, M., Vidal, J. C., and Mu-
cientes, M. (2013). Learning analytics framework for
educational virtual worlds. Procedia Computer Sci-
ence, 25:443–447.
G
¨
unther, C. W. and Rozinat, A. (2012). Disco: Discover
your processes. 940:40–44.
Gurusamy, K. S., Aggarwal, R., Palanivelu, L., and David-
son, B. R. (2009). Virtual reality training for surgical
trainees in laparoscopic surgery. Cochrane database
of systematic reviews, (1).
Harms, P. (2019). Automated usability evaluation of vir-
tual reality applications. ACM Trans. Comput. Hum.
Interact., 26(3):14:1–14:36.
Liang, Z., Zhou, K., and Gao, K. (2019). Develop-
ment of virtual reality serious game for underground
rock-related hazards safety training. IEEE Access,
7:118639–118649.
Paszkiewicz, Z. (2013). Process mining techniques in con-
formance testing of inventory processes: an industrial
application. In Int. Conf. on Business Information Sys-
tems, pages 302–313. Springer.
Rold
´
an, J. J., Crespo, E., Mart
´
ın-Barrio, A., Pe
˜
na-Tapia,
E., and Barrientos, A. (2019). A training system for
industry 4.0 operators in complex assemblies based
on virtual reality and process mining. Robotics and
computer-integrated manufacturing, 59:305–316.
Shen, H., Zhang, J., Yang, B., and Jia, B. (2019). Develop-
ment of an educational virtual reality training system
for marine engineers. Comp. Applic. in Engineering
Education, 27(3):580–602.
Vidani, A. C. and Chittaro, L. (2009). Using a task model-
ing formalism in the design of serious games for emer-
gency medical procedures. In Rebolledo-Mendez, G.,
Liarokapis, F., and de Freitas, S., editors, Conf. in
Games and Virtual Worlds for Serious Applications,
pages 95–102. IEEE Computer Society.
Wang, R., Yao, J., Wang, L., Liu, X., Wang, H., and Zheng,
L. (2017). A surgical training system for four medical
punctures based on virtual reality and haptic feedback.
In IEEE Symp. on 3D User Interfaces, pages 215–216.
Yigitbas, E., Karakaya, K., Jovanovikj, I., and Engels, G.
(2021). Enhancing human-in-the-loop adaptive sys-
tems through digital twins and VR interfaces. In 16th
Int. Symp. on Software Engineering for Adaptive and
Self-Managing Systems, pages 30–40. IEEE.
Zahabi, M. and Abdul Razak, A. M. (2020). Adaptive vir-
tual reality-based training: a systematic literature re-
view and framework. Virtual Reality, 24(4).
Towards Enhanced Guiding Mechanisms in VR Training Through Process Mining
159