Workplace Learning - Providing Recommendations of Experts and
Learning Resources in a Context-sensitive and Personalized Manner
An Approach for Ontology Supported Workplace Learning
Sandro Emmenegger
1
, Knut Hinkelmann
1
, Emanuele Laurenzi
1,2
, Barbara Thönssen
1
Hans Friedrich Witschel
1
and Congyu Zhang
1
1
University of Applied Sciences and Arts Northwestern Switzerland, Riggenbachstr. 16, 4600 Olten, Switzerland
2
University of Applied Sciences St. Gallen, Rosenbergstr. 59, 9001 St. Gallen, Switzerland
Keywords: Workplace Learning, Ontology Supported Learning, Personalized Learning, Recommender System, Public
Administration.
Abstract: Support of workplace learning is increasingly important as change in every form determines today's
working world in industry and public administrations alike. Adapt quickly to a new job, a new task or a new
team is a major challenge that must be dealt with ever faster. Workplace learning differs significantly from
school learning as it should be strictly aligned to business goals. In our approach we support workplace
learning by providing recommendations of experts and learning resources in a context-sensitive and
personalized manner. We utilize users' workplace environment, we consider their learning preferences and
zone of proximal development, and compare required and acquired competencies in order to issue the best
suited recommendations. Our approach is part of the European funded project Learn PAd. Applied research
method is Design Science Research. Evaluation is done in an iterative process. The recommender system
introduced here is evaluated theoretically based on user requirements and practically in an early evaluation
process conducted by the Learn PAd application partner.
1 INTRODUCTION
Change is given and an employee's working
environment, his/her tasks and duties changes
quickly and ever often. According to (Bureau of
Labour Statistics 2014) the median number of years
that wage and salary workers had been with their
current employer was 4.6 years in January 2014.
Already in 2012 Forbes has reported that, according
to a survey ninety-one percent of Millennials (born
between 1977-1997) expect to stay in a job even for
less than three years (Meister 2012). However, not
only 'job hobbing' requires (workplace) learning but
also taking over new responsibilities within an
organisation. In a survey conducted by Accenture
(2014) 91 percent of the respondents consider the
most successful employees to be those who can
adapt to the changing workplace. As pointed out by
Tynjälä (2008) workplace learning is different to
school learning as it is mostly informal in nature, as
- for example - usually no formal curriculum or
prescribed outcomes exits, emphasis is on work and
experiences, it is often performed collaboratively
and no distinction is made between knowledge and
skills. In our approach we aim to formalize
workplace learning by defining learning goals that
are related to business goals, objectives and
strategies. Competencies, required to reach the
learning goals and hence, the business goals, are
determined and described in job, respectively role
profiles. From this an employee's competence profile
is derived in which the level of acquired
competencies is reported, for example in an
objective agreement. Collaborative learning is
supported by using a wiki as learning platform.
For implementation we use a model driven
approach (De Angelis et al. 2015). That is, we
extended existing meta models, e.g. standard
notations like Business Process Model and Notation
(BPMN) (OMG 2011) and Business Motivation
Model (BMM) (OMG 2014) or created new ones,
based on standards (for example the Competency
Emmenegger, S., Hinkelmann, K., Laurenzi, E., Thönssen, B., Witschel, H. and Zhang, C.
Workplace Learning - Providing Recommendations of Experts and Learning Resources in a Context-sensitive and Personalized Manner - An Approach for Ontology Supported Workplace
Learning.
DOI: 10.5220/0005844907530763
In Proceedings of the 4th International Conference on Model-Driven Engineering and Software Development (MODELSWARD 2016), pages 753-763
ISBN: 978-989-758-168-7
Copyright
c
 2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
753

Meta Model is deduced from the European
Qualifications Framework (EQF) (European
Comission n.d.)) to model collaborative workplace
learning centred on business processes and their
context. We then transformed the models and
relations between them into an ontological
representation for machine execution. We also
transformed these models and relations into wiki
pages and links.
With this approach we are able to integrate
workplace learning deeply into daily business, i.e.
we consider a learner's context regarding tasks
he/she has to perform in business processes
combined with organizational knowledge about
his/her position in the organisation and, his/her
working experience. Based on this context
information, appropriate learning objects and
learning material are determined and recommended
to the learner according to his/her learning
preferences.
Our approach is part of the European funded
project Learn PAd (cf. http://www.learnpad.eu).
Applied research method is Design Science
Research (Hevner & Chatterjee 2010),
complemented by the approach of Grüninger & Fox
(1995) for ontology design and evaluation.
Application domain is Public Administration
(PA) as this sector must support extremely complex
processes in order to provide services to citizens and
companies. According to our business partner, today
it needs up to two years of learning to become fully
operational.
In Learn PAd a learning platform is created to
support Public Administration (PA) with workplace
learning. PA's can access the platform via a wiki
interface (see Xwiki, http://www.xwiki.com/en/).
For learning information about the process and
specific task(s) a learner has to perform is displayed.
As depicted in Figure 1 in the left part of the wiki
the properties of a process task as well as data input
and output is provided to the user. In the right
context-related and personalized recommendations
are given.
We assess our approach in an iterative process as
part of the overall Learn PAd project evaluation. A
first evaluation was accomplished recently.
The paper at hand is structured as follows: In
section two we will give an overview on related
work. Then we will introduce an application
scenario to illustrate our approach (section three). In
section four we will give a specification of the
recommender system, followed by a description of
its implementation (section five). First iterations of
evolution are described in section six. We conclude
in section seven.
Figure 1: Recommender Interface.
2 LITERATURE REVIEW
In our literature review we consider research on four
aspects that are most relevant for our work:
recommenders, competency frameworks, imparting
knowledge and learning styles.
Recommenders
There is today a large agreement among researchers
that e-learning content should adapt to the learner’s
context and that learners should be guided through
learning content based on such context. The
recommendation of learning objects can be regarded
as a special case of business-process oriented
knowledge management. A wide array of
recommenders have been proposed, all of which aim
at recommending the next learning activity – very
often interaction with a learning object – to a learner
who is currently engaged with an e-learning system.
Such recommendation can be based purely on a
history of learner activities, within the same or
previous sessions. Some approaches use content-
based filtering: they recommended learning items
that have a content similar to that of learning objects
in the learner’s current session (Ghauth & Abdullah
2010), (Khribi et al. 2009). Others are based on
collaborative filtering or association rule mining
(Zaíane 2002), (Khribi et al. 2009), i.e. they
recommend objects that other learners (with similar
interests) used together with the objects from the
current history. A survey of further approaches of
this kind can be found in (Sikka et al. 2012).
Other researchers claim that – besides the current
activities of the learner – additional information is
needed to make useful recommendations:
- A profile of the learner, including existing
knowledge or skill levels, preferred learning
style and current learning goal, in order to
enable proper personalization of
recommendations (Schmidt & Winterhalter
LMCO 2016 - Special Session on Learning Modeling in Complex Organizations
754

2004), (Yu et al. 2007).
- Meta information about the learning objects,
including required previous knowledge,
content type and interactivity level in order to
match them against the learner profile – and
enriched with knowledge from a domain
ontology (Schmidt & Winterhalter 2004), (Yu
et al. 2007).
- Information about the role of the learner and
his/her position in the organization (Abecker et
al. 1998, 2000), (Schmidt & Winterhalter
2004).
- Explicit information about the work context of
the learner in terms of e.g. a currently executed
task or business process (Abecker et al. 1998,
2000), (Schmidt & Winterhalter 2004).
The approaches mentioned above all use ontologies
to model the required information and rely on the
computation of similarities between a learner’s
profile (and possibly work context) and the metadata
provided with learning objects. (Yu et al. 2007)
additionally use the dependencies between learning
objects to create a “learning path” through all
recommended learning objects.
Our approach is similar to the one in (Schmidt &
Winterhalter 2004), which relies on semantic
modeling as described in (Abecker et al. 2000). We
propose to model and use the same kind of
information – i.e. we believe that all of the above
listed information is indeed necessary to make
didactically useful recommendations. We take that
approach further by concretising the meta models
and ontologies required for modelling that
information and by proposing concrete matching
procedures.
Competency Frameworks
In order to develop an appropriate competency
model we carefully studied frameworks related to
competency, like the RDCEO (The Reusable
Definition of Competency or Educational
Objective), TRACE (TRAnsparent Competences in
Europe), DeSeCo (The Definition and Selection of
Competencies) (Rychen & Salganik 2003),
DIGCOMP (Developing and Understanding Digital
Competence in Europe) (Ferrari 2013), e-CF (Anon
n.d.), Bloom's Taxonomy (Forehand 2012) and EQF
(The European Qualifications Framework)
(European Commission n.d.).
Since our application partner in the Learn PAd
project already uses the EQF framework, we decided
to base the competency model on it.
The European Qualifications Framework (EQF)
is envisaged as a meta-framework that allows
positioning and comparing qualifications. It consists
of eight reference levels which are described in
terms of learning outcomes: knowledge, skills and
competences. For instance EQF level 4 for
knowledge is "Factual and theoretical knowledge in
broad contexts within a field of work or study"; for
skill is "A range of cognitive and practical skills
required to generate solutions to specific problems in
a field of work or study"; and finally for
competence: "Exercise self-management within the
guidelines of work or study contexts that are usually
predictable, but are subject to change; supervise the
routine work of others, taking some responsibility
for the evaluation and improvement of work or study
activities" (European Commission n.d.).
Imparting of Knowledge
One of the most important aspects imparting
knowledge is the notion of a Zone of Proximal
Development (ZPD), introduced by Vygotsky
(1978). He defined the zone of proximal
development (ZPD) as "the distance between the
actual developmental level as determined by
independent problem solving and the level of
potential development as determined through
problem solving under adult guidance, or in
collaboration with more capable peers" (Vygotsky,
1978, p. 86). Vygotsky proofed that when a learner
is in the ZPD for a particular task he is able to
achieve it if appropriate assistance is provided.
Another important aspect imparting knowledge
is scaffolding. Scaffolding was coined by (Wood et
al. 1976) whose conceptualization of scaffolding
was consistent with Vygotsky’s model of instruction
and emphasizes the teacher’s role as a more
knowledgeable learner to help learners to solve
problem-oriented tasks (Kim & Hannafin 2011).
Quintana et al. stated, “the process by which a
teacher or more knowledgeable peer provides
assistance that enables learners to succeed in
problems that would otherwise be too difficult”
(2004). However, in workplace learning experts’
involvement is not always feasible. As shown by
(Boud 2003) one limitations of workplaces as
learning environments is the “reluctance by experts
to guide and provide close interactions with
learners”. Hence, other learning aids - i.e. learning
material created with certain didactic considerations
in mind, is to be recommended to support learners.
A rather young learning theory that builds also
on the ZPD idea and that takes into account the role
of technology for learning is the so-called
Workplace Learning - Providing Recommendations of Experts and Learning Resources in a Context-sensitive and Personalized Manner - An
Approach for Ontology Supported Workplace Learning
755

connectivism (Siemens, 2005). Connectivism
postulates that learning occurs when connections are
made between nodes in a learner's network - where a
node can be anything ranging from a piece of
knowledge in the learner's mind to a digital artefact
or another person. This implies that new knowledge
must be connected to existing knowledge or
experiences – which can be understood as a
concretization of the ZPD – and that such
connection can be mediated by links in the digital
environment.
Learning Styles
The theory of learning styles describes a number of
ways in which learning can be different between
individuals and claims that hence, different ways for
supporting individual learning must be developed
and adapted to a learner's individual preferences.
The Dunn & Dunn learning style model (1978)
describes several elements of learning styles: the
environmental domain, the emotional domain, the
sociological domain, the physiological domain and
the psychological domain. People deal with
information and ideas in different ways because of
their preference. These learning styles influence the
achievement of the learners. Using the right
combination of learning preferences will help the
learners to achieve their learning goals. An example
of how e-learning systems can support these
different learning styles is amongst others provided
by Wolf (2002).
3 APPLICATION SCENARIO
The application scenario was developed based on a
real case and as a result of several interviews
conducted with representatives of our application
partner in Italy, the Marche Region.
The application scenario provides all information
needed to instantiate all kinds of meta model
relevant for workplace learning, i.e. process models,
business motivation model, organisational model,
document model and competency model. We also
introduced two personas: a PA officer called
Barnaby, who joined the Public Administration of
Monti Azzurri not long ago; and an entrepreneur,
who requests a service from the PA, called Susan.
In our illustration we will focus on a complex
task of business process Barnaby is about to perform
and will show, what Barnaby must learn and how
our approach supports him.
The business process, “Titolo Unico” that
Barnaby performs provides a service to companies
who want to start a business. The process can
become rather complex and one of the most
challenging tasks is the one about involving other
PAs or private parties for contribution. To decide,
who is to involve, declarations made in the
entrepreneur's application must be carefully
assessed. We use the term Public Administration
(PA) to refer to those public administrations that
hold an office dealing with such kind of process. A
PA can be a single municipality or span a couple of
municipalities providing a service together.
The task of identifying the appropriate
organisational units to be involved while comply
with the time constraints and taking the right follow-
up decisions is of crucial importance for successfully
delivering the service. That isn't as easy as it sounds
as it requires comprehensive knowledge of the
Italian law (i.e. national, regional, provincial and
municipal norms and regulations) and, what is even
more important: much experience. Since PAs can
vary largely regarding number of organisational
units and hence specialisation due to the size of a
city or region and the nature of the PA, i.e. single or
aggregated municipalities, and the law does not go
so in detail to specify which organisational units to
involve in a particular case, experience matter a lot.
Thus, an experienced PA officer knows the law
AND the structure of the municipality to be involved
AND the responsible officers in the respective
organisational unit. Since direct contact may speed
up a task (e.g. quicker responses to requests and less
bureaucracy) this knowledge - although informal - is
highly relevant.
Learning Support
In the business process, our application scenario is
about, the entrepreneur Susan requests approval of
building a chalet on the lake of Caccamo, which
belongs to the municipality of Serrapetrona which is
in the province of Macerata, Italy. Susan uses the
application form provided at web-side of the PA and
we assume that she filled it out correctly.
By submitting the form the business process at
the PA of Monti Azzurri was started. The PA officer
Barnaby took over the task to assess the form. Based
on the type of request specific actions are to be taken
In our case the type of request is receptive tourism;
and Barnaby knows that type requires always the
authorization of the municipality according to the
Italian law (norm 9 of 2006). However, due to his
little experience, Barnaby does not know the
LMCO 2016 - Special Session on Learning Modeling in Complex Organizations
756

municipality of Serrapetrona and he is not sure of
which organisational units should be involved. He
needs an expert to advise him.
Recommending Experts:
Barnaby enters the Learn PAd system, moves to
the task “Identify Organisational Units” he has to
perform and checks on the recommendation panel
for help (see the right-hand side of Figure 1). In the
panel contact details of two experts – Sarah Brown
and Laura Cruciani - are displayed Sarah is a former
PA officer of Monti Azzurri who now works for the
municipality of Sarnano. The recommendation
system still considers Sarah as an expert as she dealt
with many cases concerning the municipality of
Serrapetrona. Laura, is the boss of Barnaby, working
for the PA of Monti Azzurri for many years.
Instead of searching internal phone books, asking
around or applying the trial-and-error method
Barnaby can contact one of the experts, who will
suggest which organisational units to involve and to
which law article it may refer. Additionally the
contact details of the personnel could also be
provided to start establishing a not too formal
business relationship.
Recommending Learning Resources:
After Barnaby got advice which organisational
units to involve, he sends requests to obtain the
opinion on the case of the involved parties.
Responses are expected within 30 days.
However, Barnaby receives answers in time from
all but one of the parties. Now he needs help in how
dealing with this situation. The Learn PAd system
has a section in the recommendation panel that
refers to learning objects and learning material (see
Figure 1). Basically all models represented in the
wiki are considered learning objects since the learner
needs to get familiar not only with a process, its
structure and tasks but also with the involved roles,
organizational units, business documents, IT systems
and so on. For differentiation we call dedicated
technical books, tutorials, learning audio and video
file etc. 'learning material'.
Thus, Barnaby checks on the learning material
provided by the Learn PAd system. As
recommendations in Learn PAd are contex-sensitive
and personalized the ZPD of a learner is considered.
More in detail, Barnaby has an acquired competency
EQF level of 3 in “Manage Specific Admin
Procedure”. Learning material recommended in
Learn PAd is also related to competencies it fosters.
In our example the book “Regulation of Titolo
Unico” - is related to the same competence
(“Manage Specific Admin Procedure”) but
classified with level 4. The difference of 1 between
the competency levels is considered conform to the
ZPD of the learner. As a learner can also determine
learning preferences (in Barnaby's case it is reading)
the recommended learning material is ranked top of
the recommendations.
Since no further challenge comes to light
Barnaby finishes the assessment of the application
and finally sends the approval to Susan for realizing
her chalet on the lake of Caccamo.
4 RECOMMENDER SYSTEM
SPECIFICATION
We learned from Vygotsky and others (1978) that
mentoring is very successful in supporting
individual learning. However, particularly in
workplace learning experts might be too busy to
provide the wishful support or spending their time
with mentoring is simply too costly. Hence, an
efficient solution is needed that provides a)
alternatives, and b) guides to experts most capable of
giving advice (with respect to expert knowledge but
also regarding the Zone of Proximal Development
(ZPD) of the learner.
In our approach for recommending relevant
information supporting the user in learning we
consider three modes of learning:
- simulation (in a simulation environment a
learner can simulate to perform a business
process task)
- browsing (a user can view and navigate
through wiki pages, representing his/her
business environment like business process,
tasks, organisational charts, related documents,
etc.), and
- execution mode (using the wiki as a front end
to perform a business service; often called
learning by doing).
Furthermore we differentiate between learning
objects, learning material and experts. As all wiki
articles correlate one-to-one to model elements they
are regarded as learning objects related to these
model elements. Learning material is information
dedicated for learning, for example (training) books,
audio and video files. Simulation and browsing is
considered as interactive learning material.
Besides the characteristics of the wiki content
(derived from the meta-model and the models), the
recommender ontology also represents
characteristics of the learning material as for
example the EQF level of knowledge that is
Workplace Learning - Providing Recommendations of Experts and Learning Resources in a Context-sensitive and Personalized Manner - An
Approach for Ontology Supported Workplace Learning
757

addressed. Furthermore the ontology contains
profiles of the learner, i.e. the workers in the PA,
including his/her EQF specification, learning
preferences and individual learning goals. With this
holistic view on learners, their working environment
and organizational network it is possible to identify
relevant learning objects, learning material and
experts, appropriate for the ZPD of the learner and
according to her learning style. An example of how
the ZPD is addressed in our approach is provided in
the previous section.
Most recommendations rely on rules. The left
side of these rules (precondition) is defined in terms
of the learner’s context - i.e. his/her required and
acquired competencies (including levels) and
learning style, as well as the context and application
data of the currently executed business process. The
right side of rules (consequence) contains the
recommended material.
Basis for Recommendations
We start from the premise that in an organisation
business goals and objectives are defined. They can
be modelled in a Business Motivation Model BMM
(OMG 2014). We extended the BMM meta model
by introducing learning goals as new Course of
Action. Learning goals can be related to business
goals and strategies that support them. To achieve a
learning goal certain competencies are needed. Note,
that we use the term competency to summarize the
three learning outcomes (knowledge, skill,
competence) defined in EQF. Hence, learning goals
defined in the BMM are related to the Competency
Model in which competencies are described
according to EQF including their levels (1-8).
We further assume, that competency profiles are
set-up for organisational units, roles etc. to specify a
set of competencies required by this entity. We also
maintain competency profiles of employees which
contain the acquired set of competencies. The
difference between the required competencies, e.g.
by a role and the acquired competencies of a person
who has this role, determines the individual learning
goal. In addition we can model specific
competencies needed for example to perform certain
tasks and hence, related to an extended process
model. In this manner we can identify the
knowledge gap a learner has, the learning goals
he/she is supposed to meet and his/her learning
preference that is also captured in the learner's
competency profile.
4.2 Making Recommendations
Depending on the learning mode recommendation
differ in range. Since the more is known about the
learner's working context the better (filtered) the
recommendation. Thus, most valuable
recommendations can be provided in the execution
mode. Here the recommender system knows exactly
what task a learner is about to perform, what tasks
are already done, what decisions have been taken
during the business process so far and what
application data is relevant. In best case within the
simulation such context information can be 'faked',
i.e. instead of real data fictional data is used but
same kind of recommendations can be provided.
Less accurate recommendation can be made within
the browsing mode as the learner is free to navigate
within one or more processes. Hence no information
is available about former actions and application
data.
Currently recommendations are given regarding
experts and learning material. Future work is to
recommend also similar cases (see Section 7).
In the following we will give two examples of
how recommendations are determined.
Recommending Experts:
The difficulty in recommending experts lies in
identifying the appropriate expert. Obviously, the
choice of an expert depends on the work situation -
and hence the knowledge required - as well as on the
level of knowledge of the learner and possibly
existing relationships between the learner and the
expert.
We consider three ways to determine experts:
1. line managers from the same organisation the
learner belongs to.
2. colleagues, having (had) the same role as the
learner but having executed the very task the
most times.
3. persons, having the same role as the learner but
belonging to another PA.
In the following the recommendation of an
experienced colleague is described in detail. As
mentioned above for building the recommender we
follow the approach of Grüninger & Fox (1995) for
ontology design and evaluation.
First the informal competency question (CQ) is
provided, followed then by its transformation into a
SPARQL query.
LMCO 2016 - Special Session on Learning Modeling in Complex Organizations
758

Informal competency question:
Given a user logged in to the Learn PAd system and
the role this user has in a task
AND
some constraints regarding task (e.g. the task a
performer is about to execute) and work
experience (e.g. a performer’s work experience)
(cf. WHERE-clause of the SPARQL query)
what internal experts can be recommended (cf.
SELECT-part of the SPARQL query)?
a. rationale: the answer is used to recommend
experts from the same organisation that executed
the tasks most often.
b. decomposition: the name of the user, user is an
actor, an actor has role in the task, role is
assigned to more than one performer, performer
has task log.
Formal competency question (SPARQL query):
SELECT ?experiencedPerformerName
?email
WHERE {
{
SELECT ?experiencedPerformer
(count(?executedTaskInstance) AS
?count)
WHERE {
?taskInstance rdf:type
bpmn:Task .
?executedTaskInstance
rdf:type ?taskInstance .
?executedTaskInstance
emo:activityIsPerformedByPerform
er ?experiencedPerformer .
?currentPerformer
emo:performerHasEmailAddress
"[e-mail]" .
FILTER(?currentPerformer !=
?experiencedPerformer)
} GROUP BY
?experiencedPerformer
}
?experiencedPerformer rdfs:label
?experiencedPerformerName .
?experiencedPerformer
emo:performerRepresentsPerson
?experiencedPerformerBusinessAct
or .
OPTIONAL {
?experiencedPerformerBusinessA
ctor foaf:mbox ?email .
}
} ORDER BY DESC (?count ) LIMIT 1
Result of the query is a colleague of the performer,
working in the same organisation, having the same
role and great work experience in the tasks the
performer is about to execute. In the
recommendation panel is name and contact details is
provided.
Recommending Learning Material:
For recommending appropriate learning
materials the zone of proximal development of a
learner must be considered. That is, the level of
competency that the learning material fosters should
be reasonably higher than the learner’s current level
of this competency (cf. application scenario
described above). Furthermore, the learning material
should support the learner's preferred style as, for
example, the learning material that matches his/her
preferred learning style is listed on top of the list and
the link to it is presented in bold characters. It is also
possible to completely filter out learning material
that doesn’t meet a learner’s learning style.
Informal competency question
Given a user logged in to the Learn PAd system and
her learning style
AND
some constraints regarding competencies (e.g.
acquired and required, i.e. fostered competencies
and their level) (cf. WHERE-clause of the
SPARQL query)
what information material is recommended? (cf.
SELECT-part of the SPARQL query)?
a. rationale: the answer is used to provide learning
material (i.e. links to documents, video files,
simulation) that are relevant to the learner, i.e.
fosters one or more competencies she has to
improve and the level of the fostered competency
is exactly one level higher than the level of the
acquired competency.
b. decomposition: the name of the user, user is an
actor, an actor has a profile, profile contains
acquired competencies and their level and the
user’s learning style, learning material, learning
material fosters one or more competency at a
certain level suitable for a certain learning style.
Formal competency question (SPARQL query):
SELECT ?learningMaterialTitle
?learningMaterialType
?learningMaterialURI
WHERE {
{
SELECT
?nextCompetencyLevelNumber
Workplace Learning - Providing Recommendations of Experts and Learning Resources in a Context-sensitive and Personalized Manner - An
Approach for Ontology Supported Workplace Learning
759

?aquiredCompetencyLabel
?learningStyle
?competencyProfile
emo:competencyProfileIsAquiredBy
Performer ?performer .
?competencyProfile
cmm:competencyProfileContainsCom
petencySet ?aquiredCompetencySet
.
?aquiredCompetency
cmm:competencyBelongsToCompetenc
ySet ?aquiredCompetencySet .
?aquiredCompetency
cmm:competencyHasLevel
?competencyLevelNumber .
?aquiredCompetency rdfs:label
?aquiredCompetencyLabel .
BIND(?competencyLevelNumber+1
AS ?nextCompetencyLevelNumber) .
?competencyProfile
lpd:competencyProfilePrefersLear
ningStyle ?learningStyle .
}
?nextCompetency
cmm:competencyHasLevel
?nextCompetencyLevelNumber .
?nextCompetency rdfs:label
?aquiredCompetencyLabel .
?nextCompetency
lpd:proposedLearningDocument
?learningDocument .
?learningDocument
elements:documentHasType
?documentType .
?learningStyle
lpd:learningStyleBelongsToDocume
ntType ?documentType .
?learningDocument
emo:documentRepresentsdocument
?foafDocument .
?foafDocument
elements:documentHasTitle
?learningMaterialTitle .
?foafDocument
eo:documentHasStorage
?documentNode .
OPTIONAL{
?documentNode
lpd:xwikiPageRepresentsNode
?learningMaterialURI .
}
OPTIONAL{
NOT EXISTS(?documentNode
lpd:xwikiPageRepresentsNode
?learningMaterialURI)
. ?foafDocument
elements:documentHasSource
?learningMaterialURI) .
}
}
After giving two detailed examples of how we build
recommendations we describe the technical
implementation of our approach.
5 RECOMMENDER SYSTEM
IMPLEMENTATION
The recommender system is an integrated part of the
Learn PAd system platform and incorporates mainly
the modelling environments, the transformation
component, the learning platform Wiki frontend and
the ontology recommender component.
The core implementation part of the
recommender system is the ontology and
recommender (OR) component. The platform
independent meta models and the conceptual meta
models are represented in OWL (Bechhofer et al.
2004) and loaded at runtime by the OR component.
The component is written in Java uses the open
source library Jena (Dickinson 2009) which provides
an API to work with ontologies.
A new set of models published via the modelling
environment will be exported in a proprietary XML
format. This exported of models are transformed in a
generic way into Wiki page representations based on
the Eclipse Modelling Framework (EMF) (Eclipse
Foundation n.d.). The transformation into the
ontology instances is using XSLT (W3C Working
Group 1999) templates and an XSLT Engine. This
approach has been chosen in a first prototype
version since it allows a straightforward
transformation directly into the specific target model
and format of the ontology. The models are
transformed into RDFS (W3C 2014) conform
classes and are formatted in the Turtle format for a
convenient work with text based version control
systems.
In a second step a more generic meta - meta model
based transformation will be evaluated.
After the transformation into the ontology, an
inferencing step is applied to run SPIN (W3C n.d.)
rules and infer relations to corresponding conceptual
model classes and eventually already existing
instance. Examples of such existing instances might
be an organisation's employee directory received
from a human resource system. The combination of
LMCO 2016 - Special Session on Learning Modeling in Complex Organizations
760

Figure 2: Ontology Levels and Transformations.
well as the transformed model objects build the
upper two levels in our OR component
knowledgebase shown in Figure 2.
Valuable recommendation rules require context
information besides the information from the
enterprise models. Application data and logging
information from process executions could provide
such information. This extended information shall be
made available for reasoning together with the
ontology and model instances. But here, we face the
problem of the missing support of multilayer
ontologies by the ontology description standards,
like OWL. If we add execution data to our ontology
we have an instance of an instance problem, i.e. the
execution data represents one layer, the process and
other model instance the next higher layer and our
PIMM/LCMM meta-models the highest layer.
Fanesi (2015) and Fanesi et al. (2015) propose an
approach with RDFS-FA respectively OWL-FA to
overcome that problem and still keep it decidable by
reasoners. Executed processes and tasks in our
example are added as instances of the process
instances. This allows applying a counting rule
which suggests a performer as an expert, if the
performer has executed the task most often.
6 EVALUATION
Before proposing the design of our
recommender, we compiled requirements based on
a) literature (see Section 2) and b) the results of a
questionnaire that was filled in by 52 civil servants.
In this section, we present a summary of how our
recommender design satisfies these requirements.
This is followed by a summary of results from a
qualitative evaluation.
Requirements Met
Regarding the interplay of the recommender with
the platform that handles the execution of the
business process and the necessary context
awareness, the following requirements were
satisfied:
- Questionnaire respondents had stated that, while
receiving recommendations on a particular task,
these recommendations should be detailed, but at
the same time they would like to keep an
overview of the whole process. This is satisfied
by presenting a process overview in the main
window of the prototype and displaying
recommendations within a sidebar.
- Civil servants emphasized that they often do not
know where the information contained in
existing or new (learning) material should be
applied. The recommender helps them in this
because recommendations are context-specific
(i.e. they get the recommendation where they
need it). Context-sensitive recommendations are
Workplace Learning - Providing Recommendations of Experts and Learning Resources in a Context-sensitive and Personalized Manner - An
Approach for Ontology Supported Workplace Learning
761

enabled by rules whose conditions are matched
to the learner’s current work context
Furthermore, requirements regarding the
competence-awareness of the recommender are
satisfied as follows:
- The choice to use EQF for the definition of
learners’ competence levels resulted in the
adoption of an EQF-based meta model for
modeling learner profiles
- Based on the definition of the zone of proximal
development (ZPD) in (Vygotsky 1978), we
formulated the requirement that the
recommender should recommend learning
objects aiming to teach the learner competencies
at a level just above her current level. This is
satisfied by describing learning objects with
intended outcomes in terms of EQF competency
levels and making sure that this level is just
above the learner’s current EQF competence
level for each recommended learning object.
Another category of satisfied requirements
concerned the adaptation of recommendations to the
learner’s learning style:
- Since questionnaire participants expressed the
desire to get recommendations for a diverse
range of content types, the recommender is able
to suggest not only documents or multimedia
learning objects, but also experts (see below) and
historical cases.
- Based on the concepts proposed by connectionist
learning (Siemens 2005) which imply the need to
make connections with a learner’s existing
knowledge, the recommender creates such
connections e.g. by proposing historical cases.
Finally, requirements regarding expert guidance are
satisfied as follows:
- Since questionnaire participants stated the need
to have quick access to recommended experts,
the recommendations include contact
information.
- Based on the notion of ZPD (Vygotsky 1978)
and scaffolding learning (Wood et al. 1976), we
ensured that recommended experts have more
advanced level of knowledge than the learner by
making rules dependent on experts’ EQF
competence levels.
Qualitative Evaluation
The qualitative evaluation consisted in a workshop
where civil servants interacted with a prototype of
the Learn PAd collaborative platform, which
included – among other functionality – the features
of the recommender. The interaction was performed
along the application scenario described in Section 3
and the corresponding application data and learner
context were known to the system. The
recommender was integrated into the prototype in
the form of a sidebar where context-dependent
suggestions were displayed.
Most of participants’ feedback revolved around
aspects of the recommender that were not yet
implemented in the prototype. Thus, participants
commented that there should be:
a) a registration form where a user’s
competencies can be assessed and then stored
in a profile.
b) more recommendations of multimedia content.
c) recommendations also on the level of the
whole process.
We consider this feedback as a confirmation that
these features will be perceived as useful when
implemented later.
7 CONCLUSION & FUTURE
WORK
With our approach we could show how workplace
learning can be improved by providing context-
sensitive and personalized recommendations for
learning in a collaborative environment. Next we
will extend recommendations to similar cases. That
is we will implement a Case Based Reasoning
System to identify and recommend cases, similar to
the one a learner is about to perform but have been
accomplished earlier.
Furthermore we will work on key performance
indicators for learning goals in order to assess
learning progress. We intend to develop a cockpit to
identify for example goals that are not satisfied and
the reasons that causes this effect.
ACKNOWLEDGEMENTS
This work is supported by the European Union FP7
ICT objective, through the Learn PAd Project with
Contract No. 619583.
REFERENCES
Abecker, A. et al., 2000. Context-Aware, Proactive
Delivery of Task-Specific Information: The
KnowMore Project.
LMCO 2016 - Special Session on Learning Modeling in Complex Organizations
762

Abecker, A. et al., 1998. Toward a Well-Founded
Technology for Organizational Memories. IEEE
Intelligent Systems and their Applications, Vol.13(No.
3), pp.40–48.
Accenture, 2014. Career Capital 2014 Global Research
Results, Available at:
https://acnprod.accenture.com/_acnmedia/Accenture/C
onversion-Assets/DotCom/Documents/About-
Accenture/PDF/1/Accenture-IWD-2014-Research-
Career-Capital.pdf.
De Angelis, G. et al., 2015. Modelling for Learning in
Public Administrations – The Learn PAd approach. In
Domain-Specific Conceptual Modelling: Concepts,
Methods, and Tools. Springer.
Anon, European e-Competence Framework (e-CF).
Available at: http://www.ecompetences.eu/e-cf-
overview (Accessed November 25, 2015).
Bechhofer, S. et al., 2004. OWL Web Ontology Language
Reference. , (February 2004).
Boud, D., 2003. Current issues and new agendas in
workplace learning. Management.
Bureau of Labour Statistics, 2014. T 2014,
Dickinson, I., 2009. Jena Ontology API.
Eclipse Foundation, Eclipse Modeling Framework (EMF).
Available at: https://eclipse.org/modeling/emf
(Accessed November 25, 2015).
European Commission, European Qualification
Framework (EQF). Available at:
https://ec.europa.eu/ploteus/content/descriptors-page
(Accessed November 25, 2015).
Fanesi, D., 2015. A Multilayer Ontology to represent
Business Process Models and Execution Data.
Fanesi, D., Cacciagrano, D.R. & Hinkelmann, K., 2015.
Semantic Business Process Representation to enhance
the degree of BPM mechanization an ontology.
Ferrari, A., 2013. DIGCOMP: A framework for
developing and understanding digital competence in
Europe. Y. Punie & B. N. Brečko, eds.,
Forehand, M., 2012. Bloom’s Taxonomy. Emerging
Perspectives on Learning, Teaching, and Technology,
p.12. Available at:
http://projects.coe.uga.edu/epltt/index.php?title=Bloo
m’s_Taxonomy.
Ghauth, K.I. & Abdullah, N.A., 2010. The Effect of
Incorporating Good Learners’ Ratings in e-Learning
Content-Based Recommender System. Educational
Technology & Society, 14(2), pp.248–257.
Hevner, A. & Chatterjee, S., 2010. Design Research in
Information Systems Integrated., Springer.
Khribi, M.K., Jemni, M. & Nasraoui, O., 2009. Automatic
Recommendations for E-Learning Personalization
Based on Web Usage Mining Techniques and
Information Retrieval. Educational Technology &
Society, 12(4), pp.30–42.
Kim, M.C. & Hannafin, M.J., 2011. Scaffolding problem
solving in technology-enhanced learning environments
(TELEs): Bridging research and theory with practice.
Computers and Education, 56(2), pp.403–417.
Meister, J., 2012. Job Hopping Is the “New Normal” for
Millennials: Three Ways to Prevent a Human
Resource Nightmare. Forbes Leadership. Available at:
hthttp://www.forbes.com/sites/jeannemeister/2012/08/
14/job-hopping-is-the-new-normal-for-millennials-
three-ways-to-prevent-a-human-resource-nightmare/
(Accessed November 16, 2015).
OMG, 2014. Business Motivation Model. BMM 1.2,
(May), p./. Available at:
http://www.omg.org/spec/BMM/1.2/PDF/.
OMG, 2011. Business Process Model and Notation
(BPMN V 2.0). Business, (January). Available at:
http://www.omg.org/spec/BPMN/2.0.
Quintana, C. et al., 2004. A Scaffolding Design
Framework for Software to Support Science Inquiry.
Journal of the Learning Sciences, pp.37–41.
Rychen, D.S. & Salganik, L.H., 2003. Key competencies
for a successful life and a well-functioning society.
OECD Definition and Selection of Competencies
Final Report, pp.1–20.
Schmidt, A. & Winterhalter, C., 2004. User context aware
delivery of e-learning material: Approach and
architecture. Journal of Universal Computer Science,
10(1), pp.28–36.
Siemens, G., 2005. Connectivism: A Learning Theory for
the Digital Age. International Journal of Instructional
Technology and Distance Learning, 2(1).
Sikka, R., Dhankhar, A. & Rana, C., 2012. A Survey
Paper on E-Learning Recommender System.
International Journal of Computer Applications, 47(9),
pp.27–30.
Tynjälä, P., 2008. Perspectives into learning at the
workplace. Educational Research Review, 3(2),
pp.130–154.
Vygotsky, L.S., 1978. Mind in society: The development
of higher psychological processes. Mind in Society
The Development of Higher Psychological Processes,
Mind in So, p.159. Available at:
http://www.amazon.com/dp/0674576292.
W3C, 2014. RDF Schema 1.1. Available at:
http://www.w3.org/TR/rdf-schema/ [Accessed
November 25, 2015].
W3C, SPIN SPARQL Inferencing Notation. Available at:
http://spinrdf.org/ [Accessed November 25, 2015].
W3C Working Group, 1999. XSL Transformations
(XSLT) Version 1.0. W3C Recommendation 16
November 1999. Available at:
http://www.w3.org/TR/xslt (Accessed November 25,
2015).
Wood, D., Bruner, J.S. & Ross, G., 1976. The role of
tutoring in problem solving. Journal of child
psychology and psychiatry, and allied disciplines,
17(2), pp.89–100.
Yu, Z. et al., 2007. Ontology-based semantic
recommendation for context-aware e-learning. In
Proceedings of the 4th international conference on
Ubiquitous Intelligence and Computing. Springer-
Verlag, pp. 898–907.
Zaíane, O.R., 2002. Building a Recommender Agent for e-
Learning Systems. In Proceedings of the International
Conference on Computers in Education. IEEE
Computer Society, p. 55.
Workplace Learning - Providing Recommendations of Experts and Learning Resources in a Context-sensitive and Personalized Manner - An
Approach for Ontology Supported Workplace Learning
763