APOSDLE: learn@work with Semantic Web Technology
(FBK-irst. Via Sommarive, 18. 38100 Trento, Italy
(Know-Center and Graz University of Technology, Knowledge Management Institute
Inffeldgasse 21a, 8010 Graz, Austria
(Know-Center and Graz University of Technology, Knowledge Management Institute
Inffeldgasse 21a, 8010 Graz, Austria
(FBK-irst. Via Sommarive, 18. 38100 Trento, Italy
(Know-Center, Inffeldgasse 21a, 8010 Graz, Austria
Abstract: The EU project APOSDLE focuses on work-integrated learning. Among the several
challenges of the project, a crucial role is played by the system’s ability to start from the context of
the immediate work of a user, establish her missing competencies and learning needs and suggest
on-the-ﬂy and appropriate learning stimuli. These learning stimuli are created from a variety of
resources (documents, videos, expert proﬁles, and so on) already stored in the workplace and
may be in the form of learning material or suggestions to contact experts and / or colleagues.
To address this challenge requires the capability of building a system which is able ﬁnd, choose,
share, and combine a variety of knowledge, evolving content and resources in an automatic and
effective manner. The implementation of this capability requires technology which goes beyond
traditional query-answering and keyword based search engines, and Semantic Web technology
was chosen by the consortium as the most appropriate technology to make information search and
data integration more efﬁcient. The aim of this paper is to give an overview of the broad spectrum
of Semantic Web technologies that are needed for a complex application like APOSDLE, and the
challenges for the Semantic Web community that have appeared along the way.
Key Words: work integrated learning, semantic web technology, APOSDLE project
Category: J.4, I.2.4, I.2.6, I.2.11
The EU project APOSDLE1focuses on work-integrated learning. In a nutshell, the
project aim is to develop a software platform and tools to support the process of learn-
Proceedings of I-MEDIA ’07 and I-SEMANTICS ’07
Graz, Austria, September 5-7, 2007
ing@work, that is learning within the context of the immediate work and current work
environment of a, so called, knowledge worker.
To deliver a knowledge worker with context-sensitive learning material, tailored to
her speciﬁc competences, work situation and learning needs, the APOSDLE system
needs to know and manipulate, not only the speciﬁc domain of knowledge in which
the knowledge worker is acting, but a set of other different aspects, spanning from
processes, competences, learning needs and methods, user proﬁles, to the knowledge
capital (documents, videos, expert proﬁles, and so on) available in the work environ-
ment and used by the APOSDLE system to build the learning material. Thus, to fulﬁl
its goals, the ASPODLE system must be able to ﬁnd, choose, share, and combine a va-
riety of knowledge and knowledge artefacts in an automatic and effective manner. The
implementation of this capability requires technology which goes beyond traditional
query-answering and keyword based search engines, and Semantic Web technology
was chosen by the consortium as the most appropriate technology to make information
search and data integration more efﬁcient. The goal of this paper is to give an overview
of the Semantic Web technologies that are needed for a complex application like APOS-
DLE, and the challenges for the Semantic Web community that have appeared along the
way. In particular we focus on a core set of functionalities that were realised as part of
the ﬁrst prototype of the APOSDLE system, and which constitute the starting point for
future enhancements. This core set of functionalities allows starting from a description
of the current proﬁle of a knowledge worker, and from her current working task, to
determine her missing competences on speciﬁc domain elements, to select appropri-
ate material from the knowledge capital of the company and compose it in appropriate
learning material. This learning material is ﬁnally presented to the knowledge worker
to help her acquire the (missing) competences required to fulﬁl her task. The realisation
of this core set of functionalities has required the capability to:
1. store an integrated representation of different domains of knowledge. This inte-
grated representation is what we call the APOSDLE knowledge base;
2. connect the knowledge capital with the APOSDLE knowledge base; and
3. retrieve suitable knowledge capital in order to compose learning material.
In the remaining sections we illustrate the Semantic Web technology and challenges
related to these three steps. We end the paper with a brief overview of the new technol-
ogy that we aim at developing for the next version of APOSDLE.
2 The APOSDLE Knowledge Base
The APOSDLE knowledge base is composed of an integrated representation of the
–Domain Model. The Domain Model is used to provide a conceptualization of the
domain-dependent knowledge with which the knowledge worker (learner) is con-
cerned, and to provide a vocabulary form the annotation of the documents which for
C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ... 263
the knowledge capital of the organization. Our choice is to represent the Domain
Model as an OWL ontology.
–Task Model. The Task Model represents the description of processes with which
the knowledge worker is concerned. Since the main emphasis of the Task Model is
to represent processes, and their dynamic aspects, our choice is to model processes
using the workﬂow language YAWL (Yet Another Workﬂow Language) and its
modelling tools [van der Aalst and ter Hofstede, 2005].
–Competence Performance Model. The Competence Performance Model is con-
cerned with the description of the competences needed to perform tasks of a cer-
tain domain (see [Ley et al., 2007]). The competence performance model can be
structured in a hierarchy from the more general competence to the more speciﬁc
competence, and a natural implementation of such a hierarchy and the relation with
tasks is via OWL.
–Instructional Model. The Instructional Model provides a conceptualization of the
main concepts used to characterise the features of documents and competences
from a learning perspective. It is mainly domain-independent and its main concepts
are taken from the IMAT ontology [Barnard et al., 1999], which is represented in
OWL in the APOSDLE system.
–Knowledge Capital. The knowledge capital we considered in the initial phase of
the project are textual documents. In the following we adopt the APOSDLE termi-
nology and use the more generic term of knowledge artefact instead of document
(see [Consortium, 2006a]).
The deﬁnition of the APOSDLE knowledge base presents two important challenges,
which are illustrated in the remaining part of the section.
Building the APOSDLE Knowledge Base. The Domain Model, the Task Models, and
the Competence Performance Model, are domain dependent models which need to be
created every time the APOSDLE system is conﬁgured and deployed for a new or-
ganization. If we consider the typical application environment of APOSDLE, we can
safely assume that the organizations will not be interested in ontology engineering and
workﬂow based process representation. Their main interest lies in setting up a tool that
enhances the productivity of their work environment and offers to transform it into an
integrated work-, learn- and collaboration environment.
This fact provides the APOSDLE developers with the ﬁrst important challenge: to
be able to provide tools that offer as much support as possible to automate and simplify
the tasks of (domain) ontology engineering, workﬂow based process acquisition, and
competence performance analysis and representation. This both in terms of graphic and
easy-to-use tool interfaces, and automatic knowledge acquisition. During the ﬁrst stages
of the project we have focused mainly on the problem of supporting the Domain Model
development, and in the current version of APOSDLE we provide a domain modelling
tool implemented as a Prot´
e plugin. This tool allows us to use all the state-of-the-
art facilities of Prot´
e, and also includes additional features similar to our earlier work
described in [Scheir et al., 2006]. Among these features are e.g. relevant term extraction
264 C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ...
and document clustering which are starting points for helping the APOSDLE ontology
engineer to elicit knowledge relevant for ontology creation.
Current and future work goes into two directions. On the one hand, usability-related
issues must be addressed. To this purpose, the existing tool is being evaluated by appli-
cation partners. On the more technical side, advanced ways of supporting users in the
task of ontology creation are being researched. Among them are enhanced preparation
of relevant terms (e.g. semantic grouping of terms using WordNet senses) and improve-
ment of document clusters that are presented to the user. Concerning the preparation of
relevant terms we are investigating the integration of the current Prot´
e plugin with
tools that allows the automatic construction of ontology out of semi-ontological struc-
tures, as for instance: concept hierarchies, classiﬁcations, ﬁle system structure, database
schemata. A detailed description of this approach is contained in [Bouquet et al., 2006].
Roughly speaking terms from semi-ontological structures are associated with WordNet
senses, and general knowledge extracted from the hierarchical structure of senses in
WordNet is used, together with possibly existing speciﬁc domain knowledge to sup-
port the phase of ontology construction. Other features may encompass guidelines to
ontology creation, guided tours for ontology creation, support for selection of suitable
ontologies from publicly available sources and embedded ontology evaluation. Our ﬁrst
attempt to embedded ontology evaluation is described in [Pammer et al., 2006].
Diverse sources of knowledge working together. The creation of the APSODLE knowl-
edge base means that the models listed at the beginning of the section need to be inte-
grated. This poses us with two problems: the ﬁrst one is how to integrate such diverse
models, especially OWL models and YAWL models, into a single structure. The second
one is how to support the (semi) automatic integration of the different models.
During the ﬁrst stages of the project we have focused mainly on the ﬁrst problem.
In particular we have decided to store the models in APOSDLE in their original for-
mat; this to exploit the peculiarities of each representational approach. In addition we
have decided to integrate them by wrapping the YAWL model in OWL and expressing
mappings between models in OWL itself. A graphical representation of the resulting
structure is given in Figure 1. An additional advantage of this approach is that we can
also access (query) the integrated models via query languages as SPARQL as if it was
a single ontology.
A next step in APOSDLE is to support the automatic semantic integration of het-
erogeneous ontology / models. We plan to achieve this by using and extending semantic
matchers like CTXMATCH [Bouquet et al., 2003].
3 Including Knowledge Artefacts into the Knowledge Base
Knowledge artefacts are connected with the APOSDLE knowledge base by means of
annotations. We have chosen to annotate them with elements of the domain ontology
and of the instructional ontology. This in order to indicate the topic(s) of a particular
knowledge artefact and its instructional value (e.g., being an Example rather than an
C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ... 265
APOSDLE knowledge base
OWL wrapping of
OWL representation of
Formal concept analysis
Instructional ontology: OWL representation
of IMAT competency type and material use
requires is of type
Knowledge artifacts annotations
is about has material use
Figure 1: The integrated view
Most of the state-of-the-art semantic annotation tools ([Reeve and Han, 2005] for a
survey) share the view that parts of documents are annotated. Within APOSDLE how-
ever, we do not need to annotate single names or small sentences of the documents.
We rather need to annotate at a paragraph or page level, since we want to use these
segments of text, here called knowledge artifacts, as basic learning elements, that is
elements that can be used to teach a concept, a procedure, and so on. Simple examples
of basic learning elements are: a self containing paragraph that provides a deﬁnition of
a concept, or a graph that describes a particular sequence of actions to achieve a goal.
It is easy to see that a typical document produced in an organization can contain
more than one basic learning element. Therefore the annotation is realized in APOS-
DLE in two main phases:
–Document segmentation. In this phase the text is segmented in parts, the knowledge
artifacts that can be used for learning purposes.
–Semantic annotation. In this phase the knowledge artefacts are labelled with a num-
ber of concepts from the APOSDLE knowledge base. A simple example: a deﬁni-
tion of ”Use Case” can be labelled with the concept ”Use Case” taken from the
domain ontology and with the tag “deﬁnition” taken from the instructional model.
Similarly to the domain modelling tool, we have implemented an annotation tool
as a Prot´
e plugin. Currently it supports manual annotation: Given an ontology, doc-
uments can be loaded into the annotation tool and semantic metadata can be assigned
manually. Additionally, the current version offers content-based classiﬁcation using the
manually annotated documents as training set.
There are multiple challenges associated with the annotation part that must be ad-
dressed in APOSDLE. First, annotation must be performed w.r.t. different models. In
266 C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ...
the current version of APOSDLE, elements from the domain ontology and the instruc-
tional ontology need to be related to documents. As a low-tech workaround, the user
could ﬁrst annotate documents using the domain model and after that proceed to anno-
tating using the instructional ontology. In the long run, this solution is hardly satisfying.
So, annotation using multiple models is necessary. The challenge hereby lies as well in
the usage of multiple models in Prot´
e as in a clear presentation to the user. Also note
that above we stated “at least one domain model”. This reﬂects the fact that we do not
suppose that we will always necessarily have exactly a single domain ontology. There
might be multiple domain ontologies that represent multiple aspects of the domain to
be learned, and this should not be a problem for annotation.
Another challenge lies in classiﬁcation of content. Currently, the integrated clas-
siﬁcation is content based (see [Scheir et al., 2006]). This is ﬁne for automatically as-
signing domain model elements to documents. However, instructional concepts like
“Explanation”, “Introduction” or “Example” may not only be deﬁned by the content
but also by the structure of a document. Finally, it will also be desirable that annotation
may be supported in the process of working with the APOSDLE system - ﬁles created
in a certain context may be automatically annotated with corresponding concepts.
4 Enhancing Text-Based Information Retrieval with Knowledge
In terms of information retrieval we see APOSDLE as application for the semantic
desktop [Sauermann et al., 2005]. We use technology for the semantic web to build a
desktop application to support the knowledge worker. In our approach we aim at com-
bining database-like queries to a knowledge representation as found in semantic web
technology with classical information retrieval approaches, i.e. the statistical analysis
of document content. We expect to be able increase the performance of our system in
terms of recall in the work support scenario and precision in the learner support scenario
with this approach.
When deﬁning a model of the context of a knowledge worker (see work described
in [Ulbrich et al., 2006]) we noticed that there are three classes of objects that can be
used to describe the current situation of the knowledge worker:
–Aset of concepts of a knowledge representation that describes the situation of the
knowledge worker, for example the current actions a person performs or the com-
petencies he or she acquires.
–Aset of documents that are related to the current situation of the knowledge worker,
for example the document template he or she is currently interacting with, or the
process documentation the person is reading.
–Aset of terms which are related to his or her current situation, examples for such
terms would be parts of documents the person currently views or a text he or she
C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ... 267
To increase the chances of successfully supporting the worker with resources, i.e. to
increase recall during a situation the person needs information to perform a certain task
we deﬁned an network model taking all three classes of objects (concepts, terms, doc-
uments) into account as query items [Scheir and Lindstaedt, 2006]. This model forms
the basis of our information retrieval system and allows for searching documents based
on content and semantic metadata (stemming from the Domain Model). Therefore we
are implementing an associative information retrieval [Crestani, 1997] system based on
the network model: in a nutshell, we create a large network (or graph) in which we com-
bine concepts, terms and documents. Starting from a given set of documents, terms and
concepts we traverse this network returning those documents most closely associated
with the given query set. As measures of associativeness we are researching semantic
and content-based similarly.
When suggesting learning material to a knowledge worker we aim at increasing
precision by again taking semantic annotations into account. This approach is different
from the one described previously but operates on top of it. The difference in the two
approaches is that when suggesting learning material we have to assure that the person
we retrieve resources for is able to learn the retrieved information. We have to assure
that the information retrieved builds up a certain competency (i.e. fosters learning) and
the person is able the execute the task without our help in the future. Therefore the pre-
sented material has to fulﬁl certain prerequisites, formally deﬁned in the Instructional
Model. For example, depending in the subject to teach we only present information of a
certain type, as certain information is learned better by giving an example, other by pro-
viding a deﬁnition. The realize this approach, we take semantic metadata of documents
(stemming from the Instructional Model) into account and ﬁlter the result set according
to the current learning situation.
In this paper we have illustrated the main Semantic Web technologies that we have
adopted in the ﬁrst stages of the APOSDLE project to support the tasks of ontology en-
gineering, ontology mapping, semantic annotation and information retrieval. This tech-
nology provides a starting point for the development of the APOSDLE platform and is
included in the description of the ﬁrst reference architecture [Consortium, 2006b]. New
technology that must be developed to make the APOSDLE platform a success concern:
(i) the ability to support user-friendly ontology engineering, (ii) the integration of dif-
ferent forms of ontologies (like the domain ontology and the competence performance
ontology) and different forms of models (like the domain ontology and the task model).
This requires matching algorithms that go beyond the identiﬁcation of equivalent con-
cepts in different ontologies; (iii) the automatic annotation of knowledge artefacts that
can be used for learning needs, and (iv) advanced algorithms for information retrieval.
268 C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ...
This work has been partially funded under grant 027023 in the IST work programme of the
European Community. The Know-Center is funded by the Austrian Competence Center pro-
gram Kplus under the auspices of the Austrian Ministry of Transport, Innovation and Technology
(www.ffg.at/index.php?cid=95) and by the State of Styria.
[Barnard et al., 1999] Barnard, Y., de Hoog, R., and Wielinga, B. (1999). Imat: Re-using multi-
media electronic technical documentation for training. In EMMSEC Conference, Stockholm.
[Bouquet et al., 2003] Bouquet, P., Seraﬁni, L., and Zanobini, S. (2003). Semantic coordination:
a new approach and an application. In Proceedings of the Second International Semantic Web
Conference, volume 2870 of LNAI, pages 130–145.
[Bouquet et al., 2006] Bouquet, P., Seraﬁni, L., Zanobini, S., and Sceffer, S. (2006). Bootstrap-
ping semantics on the web: meaning elicitation from schemas. In WWW ’06: Proceedings of
the 15th international conference on World Wide Web, pages 505–512. ACM Press.
[Consortium, 2006a] Consortium, A. (2006a). APOSDLE Scope and Boundaries - Deliverable
6.1. EU-IST Integrated Project Deliverable 6.1.
[Consortium, 2006b] Consortium, A. (2006b). APOSDLE Software Design and Development
Plan 1 - Deliverable 4.1. EU-IST Integrated Project Deliverable 4.1.
[Crestani, 1997] Crestani, F. (1997). Application of spreading activation techniques in informa-
tion retrieval. Artif. Intell. Rev., 11(6):453–482.
[Ley et al., 2007] Ley, T., Albert, D., and Lindstaedt, S. (2007). Competency management using
the competence performance approach: Modeling, assessment, validation and use. In Compe-
tencies in Organizational E-Learning: Concepts and Tools. Idea Group, Hershey, PA.
[Pammer et al., 2006] Pammer, V., Scheir, P., and Lindstaedt, S. N. (2006). Ontology coverage
check: support for evaluation in ontology engineering. In FOMI 2006. The 2nd workshop:
Formal Ontologies Meet Industry.
[Reeve and Han, 2005] Reeve, L. and Han, H. (2005). Survey of semantic annotation platforms.
In SAC ’05: Proceedings of the 2005 ACM symposium on Applied computing, pages 1634–
1638, New York, NY, USA. ACM Press.
[Sauermann et al., 2005] Sauermann, L., Bernardi, A., and Dengel, A. (2005). Overview and
outlook on the semantic desktop. In Proceedings of the 1st Workshop on The Semantic Desktop
at the ISWC 2005 Conference.
[Scheir et al., 2006] Scheir, P., Hofmair, P., Granitzer, M., and Lindstaedt, S. N. (2006). The on-
tologymapper plug-in: Supporting semantic annotation of text-documents by classiﬁcation. In
Semantic Systems From Vision to Applications - Proceedings of the SEMANTICS 2006, number
212 in email@example.com, pages 291–301. OCG.
[Scheir and Lindstaedt, 2006] Scheir, P. and Lindstaedt, S. N. (2006). A network model ap-
proach to document retrieval taking into account domain knowledge. In LWA 2006, Lernen -
Wissensentdeckung - Adaptivit¨
at, 9.-11.10.2006 in Hildesheim, number 1/2006 in Hildesheimer
Informatik-Berichte, pages 154–158. Universit¨
[Ulbrich et al., 2006] Ulbrich, A., Scheir, P., Lindstaedt, S. N., and G¨
ortz, M. (2006). A context-
model for supporting work-integrated learning. In Innovative Approaches for Learning and
Knowledge Sharing - First European Conference on Technology Enhanced Learning, EC-TEL
2006, volume 4227 of LNCS, pages 525–530. Springer.
[van der Aalst and ter Hofstede, 2005] van der Aalst, W. and ter Hofstede, A. (2005). YAWL:
Yet Another Workﬂow Language. Information Systems, 30(4):245–275.
C. Ghidini, V. Pammer, P. Scheir, L. Serafini, S. N. ... 269