Interoperability of Adaptive Learning Components
Milos Kravcik, Lora Aroyo, Peter Dolog, Geert-Jan Houben, Ambjörn Naeve,
Mikael Nilsson, Bernd Simon, Fridolin Wild
To cite this version:
Milos Kravcik, Lora Aroyo, Peter Dolog, Geert-Jan Houben, Ambjörn Naeve, et al.. Interoperability of Adaptive Learning Components. Research report of the ProLearn Network of
Excellence (IST 507310), Deliverable 1.2. 2005. <hal-00590962>
HAL Id: hal-00590962
https://hal.archives-ouvertes.fr/hal-00590962
Submitted on 5 May 2011
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
PROLEARN Deliverable 1.2
Network of Excellence Professional Learning
PROLEARN
European Sixth Framework Project
Deliverable
1.2
Interoperability of Adaptive Learning Components
Editor
Milos Kravcik
Work Package
Status
1
Document
Date
June 28, 2005
The PROLEARN Consortium
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
Universität Hannover, Learning Lab Lower Saxony (L3S), Germany
Deutsches Forschungszentrum für Künstliche Intelligenz GmbH (DFKI), Germany
Open University (OU), UK
Katholieke Universiteit Leuven (K.U.Leuven) / ARIADNE Foundation, Belgium
Fraunhofer-Gesellschaft zur Förderung der angewandten Forschung e.V. (FHG), Germany
Wirtschaftsuniversität Wien (WUW), Austria
Universität für Bodenkultur, Zentrum für Soziale Innovation (CSI), Austria
École Polytechnique Fédérale de Lausanne (EPFL), Switzerland
Eigenössische Technische Hochschule Zürich (ETHZ), Switzerland
Politecnico di Milano (POLIMI), Italy
Jožef Stefan Institute (JSI), Slovenia
Universidad Polictécnica de Madrid (UPM), Spain
Kungl. Tekniska Högskolan (KTH), Sweden
National Centre for Scientific Research “Demokritos” (NCSR), Greece
Institut National des Télécommunications (INT), France
Hautes Etudes Commerciales (HEC), France
Technische Universiteit Eindhoven (TU/e), Netherlands
Rheinisch-Westfälische Technische Hochschule Aachen (RWTH), Germany
Helsinki University of Technology (HUT), Finland
Page 1 of 16
PROLEARN Deliverable 1.2
Document Control
Title:
Interoperability of Adaptive Learning Components
Editor:
Milos Kravcik
E-mail:
Milos.Kravcik@fit.fraunhofer.de
AMENDMENT HISTORY
Version
Date
Author
Description/Comments
1 (a,b,c)
08-06-2005
Milos Kravcik
Contributions from Geert-Jan Houben, Lora
Aroyo, Milos Kravcik, Peter Dolog
2 (d)
09-06-2005
Milos Kravcik
Querying Learning Repositories from Bernd
Simon added
3 (e)
27-06-2005
Milos Kravcik
Adjusted according to the review by Stefano
Ceri, new input from Ambjörn Naeve & Mikael
Nilsson
4 (f,g)
28-06-2005
Milos Kravcik
Interoperability of Adaptive Learning added as
suggested by Stefano Ceri, input from Peter
Dolog and Fridolin Wild on SQI in adaptive
learning
5 (h)
05-07-2005
Milos Kravcik
Sections 2.1 and 5 adjusted according to the
review comments by Stefano Ceri
Contributors
Name
Company
Lora Aroyo
TU/e
Peter Dolog
L3S
Geert-Jan Houben
TU/e
Milos Kravcik
FHG
Ambjörn Naeve
KTH
Mikael Nilsson
KTH
Bernd Simon
WUW
Fridolin Wild
WUW
Page 2 of 16
PROLEARN Deliverable 1.2
Legal Notices
The information in this document is subject to change without notice.
The Members of the PROLEARN Consortium make no warranty of any kind with regard to this document,
including, but not limited to, the implied warranties of merchantability and fitness for a particular purpose.
The Members of the PROLEARN Consortium shall not be held liable for errors contained herein or direct,
indirect, special, incidental or consequential damages in connection with the furnishing, performance, or
use of this material.
Contents
1
INTRODUCTION
4
2
SEMANTIC INTEROPERABILITY OF USER-ADAPTIVE SYSTEMS
4
3
4
2.1
Applications, personalization, and user profiles
4
2.2
Distributed user profiles and semantic interoperability
5
2.3
User profile data context
5
2.4
Architecture and design
5
2.5
Representation formats and languages
6
2.6
Interoperability issues
6
FORMAL MODELS AND STANDARDS
7
3.1
Domain model
7
3.2
User model
8
3.3
Context model
8
3.4
Instruction model
9
3.5
Adaptation model
9
ACCESSING METADATA OF ADAPTIVE LEARNING SYSTEMS
4.1
Querying Learning Repositories
4.2
Querying Learner Profiles
9
9
11
5
INTEROPERABILITY OF ADAPTIVE LEARNING
13
6
CONCLUSION
14
7
REFERENCES
14
Page 3 of 16
PROLEARN Deliverable 1.2
1
Introduction
Personalized adaptive learning requires (IST, 2004) development of semantic-based and context-aware
systems to acquire, organise, personalise, share and use the knowledge embedded in web and
multimedia content, and achievement of semantic interoperability between heterogeneous information
resources and services. This includes the possibility of systems to connect to other systems in a flexible
and easy way as well as to bridge the semantical differences. To minimize the costs and effort we are not
so much interested in the possibility of interfacing that requires human effort. Semantic interoperability
can be achieved when models are more or less meant for each other and there are just some semantic
differences to detect and solve. We define model as a formal and explicit representation that can be used
to precisely describe a part of the design, e.g. the content or the user's knowledge state. Sometimes
these models also have a graphical representation which could facilitate easy communication between
designers and stakeholders, but that is not necessary.
Our task “Design and development of solutions for professional personalized adaptive learning”
includes description of interoperability for various adaptive learning components. The semantic
differences can be bridged either by standards or using approaches based on Semantic Web. This
document deals with the issue how to provide semantic interoperability of educational contents on the
web, considering the integration of Semantic Web and adaptive technologies to meet the requirements of
corporate learners. The semantic interoperability always boils down to make arrangements to transfer
data and then deal with "differences": one basic way to do this is what we describe in the first more
general part, but it is good to point out that for specific models there exist standards as we explain in the
second part. In a way one can choose to invest time beforehand by developing and choosing standards,
or afterwards – that is part of the design. The third part is considering the ways two types of metadata can
be accessed. So the structure of this deliverable is as follows: In the Introduction we have clarified the
motivation of this document. Section 2 deals with semantic interoperability for user-adaptive systems.
Section 3 presents formal models and standards. Section 4 shows how metadata of adaptive learning
systems can be accessed.
2 Semantic
Systems
Interoperability
of
User-Adaptive
The increasing demand for personalization in (e-learning) applications leads to a process of user profiling
which is inherently distributed. For applications to effectively share and exchange user information for
adaptation, they need to know the semantics of this user information, and therefore resolve the issues
related to semantic interoperability. In this part we consider the state of the art in semantic interoperability
in relation to distributed user profile in terms of methods, techniques, tools, and issues related to
semantic interoperability.
2.1
Applications, personalization, and user profiles
In the past decade we have witnessed a growing interest in applying adaptation and personalization in
numerous application domains. The process of engineering of information systems has shown a
considerable change and adaptation is a significant driver for this change. Concept-based systems
represent content using concept structures, in the sense that a model of the content (often referred to as
domain model or content model) is a characteristic element of the design. This model includes, as
relevant aspects, the user's knowledge (user model) or the adaptation knowledge (adaptation model).
These systems are distinguished from systems in which the adaptation is defined without an explicit
model of the content (e.g. because the content and structure are rather straightforward or small).
Adaptive concept-based systems are becoming especially accepted in application areas where the main
goal is to tailor large amounts of information to the individual preferences and knowledge state of the
different users. In the case of educational applications, it has become more or less standard that the
system expresses a behavior that matches the specific user (as, by the way, was long the case for the
real classroom where the teacher would approach each student differently). The construction of conceptbased systems is not a straightforward issue, certainly when the challenge is combined with the desire to
add adaptation, e.g. adaptive hypermedia systems, adaptive Web information systems, and adaptive
task-based systems. When we talk about adaptation and personalization, the user plays a fundamental
role in the system and therefore in its design. The system might want to record the user's preferences,
but also its assumption on the user's (knowledge) state. The systems typically maintain a model of the
individual user as an overlay of the domain model in order to record the current state of the user with
respect to his/her knowledge of domain concepts. The application dependent user models with the
Page 4 of 16
PROLEARN Deliverable 1.2
preferences and the state of the user are integrated in the user profile that in general comprises the
available information about a user, and is used as a basis for adaptation of the content presentation to
the user.
2.2
Distributed user profiles and semantic interoperability
The nature of cooperation between systems and applications implies that there is a distributed process of
sharing and exchanging user profiles. In order to be able to effectively manage the distributed user
profiles, and control this process of on the one hand providing profile information for personalization and
on the other hand consuming user models for personalization or adaptation, architecture for distributed
profile exchange and management is necessary. The issue of semantic interoperability in the context of
user profiling is a direct consequence of the distribution in user profile information. The decentralized
process of distributed user profile information management demands a control that is essential for a
successful application of user profiles.
2.3
User profile data context
For long it has been very difficult or even impossible to share and exchange user profile data. More
recently, suppliers and consumers of user profiles have become more aware of the need for standards
for the representation and exchange of user profile data, and especially the e-learning domain is making
enormous progress. At the same time we observe that the amount and diversity of profile-based
applications makes it practically impossible to easily create a unified "user profile infrastructure". One
important aspect that should not be underestimated is that the metadata for user profile data implies a lot
of manual labor before the metadata can effectively be exploited in the exchange of profile data between
applications. The technological advances of the last few years, especially around the Web and the
Semantic Web, can come to the rescue with a demand for tools and methods
• to combine the available data,
• to annotate profile information automatically or semi-automatically,
• to supply applications with the necessary profile metadata.
The (semi-)automatic generation of metadata is an essential prerequisite for the semantic
interoperability of profile-based applications such as e-learning applications. The creation of such
metadata usually requires a considerable intellectual input of humans. Current Web technology may offer
opportunities for semantic interoperability between applications and their metadata on a large scale,
which could not be achieved by human input alone. When we investigate how the automatic creation of
semantic metadata can be achieved, we observe that ontologies (see below) provide an option for
semantic coherence between profile data items. Tools could then minimize the amount of user effort
required for creating and maintaining semantic annotations and could thus help to increase the overall
quality level of annotations.
2.4
Architecture and design
To manage distributed user profiles, an architecture for distributed profile exchange and interpretation is
needed. Different types of systems use different kinds of architectural solutions. There are differences in
the way in which user profiles are used, and this has consequences for the personalization. As basic
architecture types, here we mention:
• adaptive Web-based systems
• adaptive hypermedia systems
• adaptive task-based systems
All these architectures share the facility to maintain a representation of assumptions about one or
more types of user characteristics in models of individual users. In other words the system should
maintain a model about the user that for instance contains assumptions about their knowledge,
misconceptions, goals, plans, preferences, tasks, or abilities. We list a number of issues related to this
user representation that need to be considered in a complex approach:
• user environment (e.g. class, school, family, background)
• roles and stereotypes
• historic and sensor information
• trust and acceptance
• generality and domain independence
• expressiveness
• inferential capabilities
• import and export
• privacy
Page 5 of 16
PROLEARN Deliverable 1.2
• mobility
Obviously there are also different ways to communicate user profile data, e.g. via a centralized
server, via peer-to-peer communication, using agent-based techniques, or using a constraint-based
approach.
2.5
Representation formats and languages
For such a distributed user profile architecture, data models and languages for profile metadata are
needed, especially to describe the semantics and semantic differences. The languages and technologies
designed for the development of the Semantic Web provide useful instruments for the representation of
semantics of profile data. We mention the concept of ontology as “an explicit specification of a
conceptualization”. This basically means that an ontology is a formal way of describing (some aspects of)
the real world. With this key concept, the Semantic Web research has given us languages that are useful
for the basic interoperability of user profile data. The Semantic Web provides a framework for expressing
and using ontologies through the use of RDF, RDF Schema and OWL. These languages come also with
relevant tool support, such as APIs, e.g. Jena and Sesame, browsers and editors, e.g. Protege and
KAON, and reasoners.
2.6
Interoperability issues
Representing the semantics of the user profile data is one step in the process, but with the distribution
come several interoperability problems and issues related to the semantics metadata. As examples, we
have incompatibility (both between profiles and between profiles and applications), incompleteness in the
sense of information missing from profiles, and contradiction in (unified) profiles.
The need to consider these issues arises from the fact that a learner may attempt to use an
application that requires more information than the user’s profile can provide, or that responds with
information that cannot be accommodated in the user’s profile. The complementary case arises when an
application cannot handle parameters such as preferences, specified by the learner or provides a
response that contains too little information to enable the user to choose between alternate follow-up
actions. Another class of problems arises when the learner’s profile contains sufficient information but the
application possesses information of its own that disagrees with the information present in the learner’s
profile, because of conflicting values or semantics.
When two e-learning applications are directly interacting to provide a learner with a certain
service, but without the direct involvement of the user, they may face the situation that they may have a
partly overlapping but not complete view of the user profile. The question now becomes how to resolve
the overlap and fill in the gap. When applications are allowed to fill in missing data themselves, it could
occur that two applications fill in contradictory data. Can and should something like this be prevented
from happening? And if not, how can the contradictory data be corrected afterwards, or how can possibly
conflicting data that co-exists simultaneously be dealt with?
Other sources of contradictory information are different versions of the same information
(freshness). The issue here is whether to trust the most recent version (newer is better in the age of cut
and paste?) or to establish a procedure to validate information. When information statements inside one
source (document) are contradictory one speaks of an inconsistency. When information statements from
different sources are contradictory, one speaks of disagreement. A disagreement may turn up when two
sources are merged (in reality in a warehouse project or virtually as above). These two situations require
different handling. In the context of interoperability, one may assume as a starting point that the sources
are consistent. The proper treatment of disagreement is the more relevant problem to tackle.
These examples illustrate the situations that have to be prepared for and dealt with. The
information related issue can be discussed from several angles: at the level of schemas or ontologies, or
at the level of instances, within an information source, or between information sources. The issue is how
to identify and deal with missing or incomplete information.
When it comes to the techniques and architectures to be considered for solutions to
interoperability, we can benefit from results from classical databases, data warehouses, mobile
information systems, and the Semantic Web. Several issues are relevant in this context and will need
solutions – here is just a brief outline:
• imprecise information
• imprecise manipulation
• uncertain information
• schema and ontology mapping
• data cleaning
• inconsistency
• mediation
Page 6 of 16
PROLEARN Deliverable 1.2
•
•
•
data dissemination
data replication
conflict detection and reconciliation
3 Formal Models and Standards
The knowledge driving the adaptation process can be represented in adaptive hypermedia systems as
five complementary models (Figure 1) – the domain model specifies what is to be adapted, the user and
context models tell according to what parameters it should be adapted, and the activity (instruction) and
adaptation models express how the adaptation should be performed. We use this model to identify the
different design aspects in which the separation between applications asks for interoperability. Note that
individual models may be distributed in reality. In the following paragraphs we discuss formal models and
standards that apply to each of the particular models. As we can see the existing standards do not really
support interoperability as a common abstract model is missing. They can be used in isolation, but this is
not desirable.
Figure 1: Enhanced Adaptive Hypermedia Application Model
3.1
Domain model
The domain model specifies the conceptual design of an adaptive hypermedia application, i.e. what will
be adapted. The information structure of a domain model in a typical adaptive hypermedia system can be
considered as two interconnected networks of objects (Brusilovsky, 2003):
• Knowledge Space – a network of concepts
• Hyperspace – a network of hyperdocuments
Accordingly, the design of an adaptive hypermedia system involves three key sub-steps:
• Structuring the knowledge
• Structuring the hyperspace
• Connecting the knowledge space and the hyperspace
3.1.1
Knowledge Space
Modern AHSs model the domain as a semantic network (Brusilovsky, 2003). They use network models
with several kinds of links that represent different kinds of relationships between concepts. The most
popular kind of links in educational AHS is prerequisite links between concepts which represent the fact
that one of the related concepts has to be learned before another. Other kinds of links that are popular in
many systems are classic semantic links “is-a” and “part-of”. These domain ontologies represent the
Page 7 of 16
PROLEARN Deliverable 1.2
expert’s knowledge about the domain. The domain model offers a natural framework for goal modeling.
An individual educational goal can be modelled as a structure (e.g. sequence, tree, stack) of subsets of
domain concepts.
3.1.2
Hyperspace
The Learning Object Metadata (LOM) standard defines a learning object as any entity, digital or nondigital, that may be used for learning, education or training (LOM, 2002). Content models identify different
kinds of learning objects and their components. A comparative analysis of six known content models
(Verbert & Duval, 2004) led to the creation of a general model that includes the existing standards and
distinguishes between:
• Content fragments – learning content elements in their most basic form (text, audio, video),
representing individual resources uncombined with any other; instances
• Content objects – sets of content fragments; abstract types
• Learning objects – they aggregate instantiated content objects and add a learning objective
The standards that can be used at this level include
• IMS Content Packaging – description and packaging of learning material
• IMS Question and Test Interoperability – XML language for describing questions and tests
• IEEE Learning Object Matadata – description of learning resources
3.1.3 Connecting Knowledge Space with Hyperspace
According to (Brusilovsky, 2003) the process of connecting domain knowledge with educational material
is also known as indexing because specifying a set of underlying concepts foe every page of educational
material is very similar to indexing a page of content with a set of keywords. There are four important
aspects to distinguish indexing approaches:
• Cardinality – single concept indexing (each fragment is related to one concept) and multi-concept
indexing (each fragment can be related to many concepts)
• Expressive power – the amount of information that can be associated with a link between a
concept and a page
• Granularity – concerns the precision of indexing (e.g. the whole page, fragments)
• Navigation – whether the link between a concept and a page exists only on a conceptual level or
also defines a navigational path
3.2
User model
The majority of educational AHS use overlay model of user knowledge (Brusilovsky, 2003). The key
principle of the overlay model is that for each domain model concept, individual user knowledge model
stores some data that is an estimation of the user knowledge level on this concept. A weighted overlay
model of user knowledge can be represented as a set of pairs “concept-value”, one pair for each domain
concept. Some systems store multiple evidences about user level of knowledge separately. Another
alternative to model the user knowledge is provided by historic model that keeps some information about
user visits to individual pages. Some AHS use this model as a secondary source of adaptation.
The learner’s goals can be modelled as a set of concepts (competencies) that can be
represented similarly to the overlay model. Additionally to these dynamic dimensions the leaner model
includes also a more static one – user preferences. The most relevant ones are preferred cognitive and
learning styles, as well as the language. The main challenges and requirements in this field include
generic user modeling, enabling reusability and sharing of the model by various applications, as well as
group modeling.
The following standards relate to user modeling:
• IEEE Public And Private Information – specifies both the syntax and semantics of a 'Learner
Model,' which will characterize a learner and his or her knowledge/abilities
• IMS Learner Information Package – learner information data exchange between systems that
support the Internet learning environment
3.3
Context model
The user (learner) and context model specify to what parameters the application should adapt. One of the
primary objectives is to generate as much metadata as possible automatically, based on the current
context and possibly by sensors (additionally to the time parameter also other suitable attributes, e.g.
GPS coordinates, temperature, etc). This will enable more precise retrieval of the data when learning
objects are processed or elaborated by students and teachers.
Page 8 of 16
PROLEARN Deliverable 1.2
Context management has to deal with such issues as automatic acquisition of context metadata,
contextualized delivery of content, contextualized delivery of activities (interaction of users), and
contextualized delivery of services. The current standards and exchange formats for contextualisation of
resources have to be extended. Designing context-based activities involving groups of users interacting
within a set of collaborative environments should be enabled. There are no standards related to the
context model yet.
3.4
Instruction model
The instruction (pedagogical) and adaptation models specify the navigational design for an adaptive
hypermedia application. Together with the presentation specification they tell how the adaptation should
be preformed, so they describe the dynamics of the system.
Learning design is a way of modeling learning activities and scenarios, as different types of
learners prefer different learning approaches – learning styles. A key axiom that is common to all major
educational approaches says that Learners perform Activities in an Environment with Resources (Koper,
2001). The IMS Learning Design uses the metaphor of a theatrical play to describe the workflow involved
in learning and teaching scenarios. Main challenges include encoding dynamic interactions between
users and system, representing scenarios (objectives, tasks/activities) and describing interactions
between participating roles and system services, as well as separation of scenarios from resources
(reusability).
Related standards:
• IMS Simple Sequencing – representing the intended behaviour of an authored learning
experience
• IMS Learning Design – defining diverse learning approaches (scenarios)
3.5
Adaptation model
This model specifies the adaptation semantics – which objects are seen, mastered, recommended, etc.
Adaptation specifications define the status of individual objects (e.g. content objects or fragments) based
on their attributes and the current parameters of the user model, or more generally of the context model.
The adaptation effect is usually achieved by adapting contents and links using suitable adaptation
techniques that can be chosen on this level. The taxonomy of adaptive hypermedia technologies
(Brusilovsky, 2001) includes:
• Adaptive presentation (content level adaptation) to ensure for different classes of users that the
(most) relevant information is shown and the user can understand it:
o Adaptive text presentation
o Adaptive multimedia presentation
o Adaptation of modality
• Adaptive navigation support (link level adaptation) to guide the user towards the relevant,
interesting information:
o Direct guidance
o Adaptive link sorting
o Adaptive link hiding
o Adaptive link annotation
o Adaptive link generation
o Map adaptation
4 Accessing Metadata of Adaptive Learning Systems
Besides the semantic interoperability, the systems must understand access mechanisms to learning
content objects, learners and associated metadata. They must know programming interfaces to connect
to, retrieve and manipulate needed metadata. The Application Program Interfaces (API) are either
domain specific, i.e. they are based on specific metadata models or are generic, usually suitable to query
metadata based on multiple schemas by making use of general purpose query languages like SQL.
4.1
Querying Learning Repositories
Interoperability among learning repositories requires a common communication framework for querying.
In the next sections we present an overview of different query APIs in the learning domain (Simon et al.,
2005) and Simple Query Interface (SQI, 2005) – an API for querying learning objects repositories. The
overall objective of these activities is to build up a global network of learning object repositories.
4.1.1
Query APIs in Learning Domain
Page 9 of 16
PROLEARN Deliverable 1.2
OpenURL (OpenURL, 2004) as well as the Content Object Repository Discovery and Resolution
Architecture – CORDRA (CORDRA, 2004) are initiatives that investigate the “Identifying” problem. The
work on SQI is “orthogonal” to this, in that queries and results can refer to identifiers of arbitrary nature.
Z39.50-International: Next Generation (ZING) covers a number of initiatives by Z39.50
implementers to make Z39.50 (ZING, 2001) more broadly available and to make Z39.50 more attractive
to information providers, developers, vendors, and users. SRW is the Search/Retrieve Web Service
protocol, which is developed within ZING and aims to integrate access to various networked resources,
and to promote interoperability between distributed databases, by providing a common utilization
framework. SRW is a web-service-based protocol (SRW, 2004). SRW takes advantage of CQL
("Common Query Language"), a powerful query language, which is a human-readable query.
SRW has many similarities with SQI, but also some differences. SRW is purely synchronous
(source-initiated), i.e. query results are returned with the response. Additional query results can be
retrieved later from the results set stored at the target for a pre-defined amount of time. SRU, the Search
and Retrieve URL Service, is a companion service to SRW, the Search and Retrieve Web Service. Its
primary difference is its access mechanism: SRU is a simple HTTP GET form of the service (SRU, 2005).
SRW encourages the use of Dublin Core, but is in general schema neutral (like SQI). SRW packs all the
functionalities in a few methods and does not adhere to the “Command-Query separation principle”. SRW
does not provide hooks for authentication and access control nor is it based on a session management
concept. It defines an Explain operation, allowing a client to easily discover the capabilities and facilities
available at a particular server. SRW uses a rich set of XML-encoded application level diagnostics for
reporting errors. SQI uses faults.
The purpose of the IMS Digital Repository Interoperability (DRI) Specification (IMS, 2003) is to
provide recommendations for the interoperation of the most common repository functions. The DRI
specification presents five core commands, i.e. search/expose, gather/expose, alert/expose, submit/store,
and request/deliver, on a highly abstract level. The specification leaves many design choices for
implementers. For example, while recommending Z39.50 (with its own query language) it also
recommends XQuery as a query language. The query service does distinguish between asynchronous
and synchronous query mode.
The EduSource project (Hatala et al., 2004) aims to implement a holistic approach to building a
network for learning repositories. As part of its communication protocol – referred to as the EduSource
Communication Language (ECL) – the IMS Digital Repository Specification was bound and implemented.
A gateway for connecting between EduSource and the NSDL initiative, as well as a federated search
connecting EduSource, EdNA and Smete serve as a first showcase.
OKI (Open Knowledge Initiative) is a development project for a flexible and open system to
support on-line training on Internet (OKI, 2004). OKI has issued specifications for a system architecture
adapted to learning management functions. One of the main characteristics of the project is its
commitment to the open source approach for software component development. OKI supplies
specifications for a model of functional architecture and an API called Open Service Interface Definition
(OSID). OKI OSID main aspects are:
• To supply specifications for a flexible and open source model of functional architecture
• Service Interface Definitions (SIDs) organize a hierarchy of packages, classes and agents and
propose Java versions of these SIDs for use in Java-based systems and also as models for other
object-oriented and service-based implementations.
• Components developed by OKI are compliant with specifications issued by IMS and ADL SCORM.
The Resource Description Framework (RDF) is one of the key pillars of the Semantic Web (RDF,
2005). RDF is an extensible way to represent information about (learning) resources. One of RDF’s
design assumptions is that resources are identified by a Unique Resource Identifier (URI) allowing
various users and agents to make assertions about uniquely identified things. RDF is designed for
representing metadata about all kinds of digital and non-digital artifacts making it a powerful means of
integration over disparate sources of information. The graph-based structures of RDF can be serialized in
XML. XHTML 2.0 is currently under development, which will support a seamless integration of RDFbased meta-tagging in HTML.
The W3C has designed SPARQL (SPARQL, 2005) as a query language for RDF. SPARQL is
designed to meet the following requirements:
• Conjunction
• Disjunction
• Optional Match
• Extensible Value Testing
• Limited Datatype Support
The development was aligned towards the following design goals:
• Human-friendly Syntax
Page 10 of 16
PROLEARN Deliverable 1.2
•
•
•
•
Data Integration and Aggregation
Non-existent Triples
Addressable Query Results
Sorting Results
Edutella is an RDF-based Peer-to-Peer infrastructure for querying distributed learning object
repositories, that comes with its own query language QEL (QEL, 2004), which is similar in functionality to
SPARQL.
4.1.3
Simple Query Interface (SQI)
SQI (Aguirre et al., 2005) is an API (developed as part of PROLEARN, see Deliverable 4.1 focusing on
SQI for details) that provides method support for asynchronous and synchronous queries. The underlying
common schema is specifically designed to the needs of an educational network of training measures
while reusing standardized concepts from IEEE LOM and Dublin Core at the same time. One of its major
design objectives was to keep the specification simple and easy to implement. The collaborative effort of
combining highly heterogeneous repositories has led to the following requirements:
• SQI is neutral in terms of results format and query languages: The repositories connecting via SQI
can be of highly heterogeneous nature: therefore, SQI makes no assumptions about the query
language or results format.
• SQI supports Synchronous and Asynchronous Queries in order to allow application of the SQI
specification in heterogeneous use cases.
• SQI supports, both, a stateful and a stateless implementation.
• SQI is based on a session management concept in order to separate authentication issues from
query management.
The design of the API itself is based on following design principles:
• Command-Query Separation Principle,
• Simple Command Set and Extensibility.
The SQI is part of a Learning Object Repository Interoperability (LORI) Framework. LORI is a
layered integration architecture, which defines services to achieve interoperability among learning
repositories. These services include core services, for example authentication service, session
management service and application services like query management or provision services. There are
already some applications of SQI in adaptive learning as we mention in the following.
Human Capital Development (HCD) Suite (http://www.hcm-online.com/ubp) is an application
especially designed to support goal-driven human capital development processes. It provides a service
for identifying and satisfying knowledge gaps and matches them with offers from different service
providers according to the needs of the company and the individual learner. It uses a ranking component
to rank search results from elena smart spaces for learning. The component assumes that the resources
in the smart space are annotated and classified by a skill ontology to be used by a user. An annotator has
been developed, facilitating extension of learning resource metadata with specific skill ontology concepts.
Annotator is based on metadata analysis and document analysis techniques to get these additional
extensions. Ontology concepts are used to index and classify metadata and content (based on term
frequency analysis). In both cases, a similarity between concepts on the one hand and metadata or
content on the other hand is computed. Highest similarities then determine which resources should be
annotated by particular concepts. The extended annotations are used for ranking purposes.
Alocom (http://memling.cs.kuleuven.ac.be/alocom/) is a framework which allows to "split" Open
Office presentation files (and Powerpoint files as well) into their building blocks to allow for retrieval of
sub-presentation objects and (semi-)automatic generation of presentations. Its specification for the
indexing interface closely relates to SQI.
4.2
Querying Learner Profiles
The Lerner API (Dolog & Schäfer, 2005) was developed in the context of FP5 EU/IST project Elena –
Creating Smart Spaces for Learning (http://www.elena-project.org). The API is based on learner ontology.
Figure 2 depicts an excerpt of a learner profile ontology configured from fragments based on three
specifications (the Elena project web site and its personalization section provide complete ontology in
RDFS). The abbreviated syntax for namespaces is used in concept and relation labels (e.g. qti stands for
Question and Test Interoperability namespace at http://www.elena-project.org/images/other/qtilite.rdfs).
The default namespace is http://www.elena-project.org/images/other/learner.rdfs.
The conceptual model describes a situation where a learning performance (IEEE PAPI is used to
model performance and portfolio, http://ltsc.ieee.org/archive/harvested-2003-10/working_groups/wg2.zip)
of a student is exchanged as his achieved competency records (IMS RDCEO – Reusable Definition of
Competency and Educational Objectives, http://www.imsglobal.org). The competencies have been
Page 11 of 16
PROLEARN Deliverable 1.2
evaluated by learner assessment (e.g. tests) and were derived from learning objectives of tests (IMS
QTI). Furthermore, all other educational activities, further materials, and projects created within the
activities are reported within the portfolio of the performance. Additional information which is reported
under preferences (IMS LIP) comprises language, device, resource and learning style preferences. The
standards and open specifications guarantee wider acceptance between e-learning systems and as such
can be seen as good candidates for the learner exchange models.
Currently, none of the referenced standards present their metadata in a way that makes it
possible to use them in combination as depicted above. Therefore, an RDF translation of these standards
had to be developed, which made it possible to use them in combination. This RDF translation is
unofficial, and we therefore view it as an important direction for future standardization work that the
standards use a common framework such as RDF and the Semantic Web, to enable the added value of
using the standards together.
Figure 2: An excerpt of a conceptual model for learner profile based on standards
Figure 3 depicts several possible scenarios of how to access and exchange learner profile
fragments. The fragments can be accessed programmatically by the use of a Java API, the web service
which exports the learner model through the API and acts as a learner model server, and through a query
infrastructure for RDF repositories like Edutella (Nejdl et al. 2002).
Exchange of Learner Profile
between Adaptive Applications
by Using API
HTTP, Use of Adaptive Application
XML-RPC/SOAP
Application
protocol based
Server
exchange of
XML-RPC/SOAP
Learner Profile protocol based
exchange of
User Model Server Learner Profile
Learner’s
Laptop
Web Service
User Model Server
Web Service
Adaptive ApplicationApplication Server
Web Service
Learner API
Edutella Network
Edutella Provider
Edutella Provider
Edutella API
Web Service
Interface
Figure 3: The use of the API in several scenarios
4.2.1
Implementation in Java
A Java API has been developed. It is structured according to the learner ontology fragments mentioned
above. The API is meant to be used to retrieve, insert, and update the learner profiles stored in the
structures described above. The API defines a class and properties for each class from the RDFS for the
learner model. The interface provides access functions for getting, deleting and updating a model of the
fragment. It provides further functions to derive additional information or to process more complex
Page 12 of 16
PROLEARN Deliverable 1.2
manipulations over referenced information types as well. The API is implemented for the RDF
representation (instances of the RDFS described above). The API is easily extensible by providing further
specializations if additional extensions and interface implementations for local repositories and data
models are needed.
4.2.2
Implementation as Web Services
The second implementation is provided through web services where several clients can access one
model which is persistent on one server. The server holds the main model, i.e. the data of a learner
profile gathered from several sources, and handles all requests from the clients. Each client is uniquely
identified at the server and can be used by a browsing or assessment system. Furthermore, a client can
be used by other learning systems which want to make use of the learner profiles or which want to
contribute to them. The model can be accessed directly by invoking functions of a web service or in a
synchronized replicated way; i.e. each client has its own repository which is synchronized with the main
server every time a change occurs. The web services framework can be used in a distributed way as well
(several servers exchanging learner models between each other).
4.2.3
Retrieval through RDF querying infrastructure
The learner profiles are created in RDF. Therefore, a query infrastructure for RDF data is another access
option. Edutella provides a datalog-based language to query RDF data provided in a distributed P2P
environment. This option enables to collect various fragments by utilizing for example the algorithm from
(Dolog, 2004). Another advantage of the P2P sharing infrastructure used with the learner profiles is that it
can facilitate an expert finding based on the provided profile which can be queried by people who need a
help in learning.
5 Interoperability of Adaptive Learning
Learning objects distributed in various repositories with associated metadata provide the opportunity of
using federated search. Early adopters have started using these services. These users can be either
learners using learning objects in a similar way like textbooks, or teachers that need suitable materials to
support their classes and possibly applying blended learning approaches.
Reuse, interoperability, and personalization belong to the main aims of IMS LD. To allow
personalization a method can contain conditions (Koper & Olivier, 2004), i.e. If-Then-Else rules that
further refine the assignment of activities and environment entities for persons and roles. Conditions can
be used to personalize LDs for specific users. The ‘If’ part of the condition uses Boolean expressions on
the properties that are defined for persons and roles in the LD. Notwithstanding IMS LD can be used to
model and annotate adaptive learning design, designing more complex adaptivity behaviour might be not
too easy. For instance, it is not possible to annotate learning content or define student roles considering
their characteristics. Moreover, it is premature to determine the reusability level of learning designs
(Koper, 2005). We agree with the finding – from the area of learning objects – that the more context is
assigned to the objects the lower is their reusability (Hodgins, 2005); such finding is valid also for learning
activities, i.e. learning design and adaptivity. Therefore it would be beneficial to distinguish well-defined
learning layers so that each object of a given layer can be substituted with other objects of the same layer
and combined with other objects at a different layer so as to build a complete solution. These solutions
can be possibly created by different authors.
In the WINDS project (Kravcik et al., 2004; Kravcik & Specht, 2004) we have experienced that
authors without programming skills can produce adaptive courses by specifying declarative knowledge
for adaptation by means of metadata like pedagogical roles of learning objects and content fragments.
This together with procedural knowledge encoded in the course player can generate adaptive delivery of
courses. We have also attempted (Kravcik, 2004) to generalize the WINDS experience with the aim to
simplify authoring as well as to achieve more flexibility, reusability and interoperability of partial learning
resources. At least on the authoring level we need separation of learning design, adaptation and
presentation specifications. Our approach is based on a recognition how personalized learning
experience is usually delivered:
1. The learning objective is specified
2. The teacher chooses a learning scenario structuring suitable learning activities
3. The teacher assigns learning objects (with the concrete learning objectives and pedagogical
roles) to specific learning activities
4. The delivery of a learning object depends on the characteristics of the particular learner, e.g.
the learning style
Page 13 of 16
PROLEARN Deliverable 1.2
5. The presentation of the leaning object depends on the current context, e.g. the delivery
device
This approach is typical when the teacher works with a fixed group of learners. Of course,
providing fully individualized learning the learner’s characteristics should be considered already in the
step 2 and 3 and the group itself can be formed according to the individual traits of learners. From the
technological point of view a major issue is what should be specified at which level to achieve reusability
and interoperability of resources. We have already specified the layers in the formal model, so let us
discuss now how it influences the authoring process.
1. Domain layer: This level includes learning objects, ontologies and metadata. Especially
important are the pedagogical metadata, like pedagogical roles. It is also important to
consider redundancy of learning object and content fragments if personalized adaptive
learning is to be provided. This means for instance that a content fragment with a certain
pedagogical role (e.g. definition, example, fact) should be created in alternative media – it
can be called learning fragment.
2. Instructional (pedagogical) layer: Learning design includes definition of learning scenarios
that can depend on the individual learner, especially her learning style. This means the
author creates a structure of learning activities that will provide the learning experience.
Examples of such learning activities are question formulation, self learning, explanation, field
trip, investigation, creation of an artefact (e.g. essay, design), artefact annotation,
performance (e.g. report, review), artefact evaluation. The teacher can provide references to
learning objects for certain learning activities, e.g. for those related to expository or
exploratory learning (that can be supported by various searching, navigation and visualization
facilities in the user interface).
3. Adaptation layer: We can distinguish adaptation strategies and adaptation techniques at this
layer. Adaptation strategies specify how to choose relevant learning objects and how to order
them, both taking into account the specific learner, especially her learning style. In this
process pedagogical roles of learning objects and other relevant pedagogical metadata
should be considered. Adaptation techniques define how to select suitable content fragments
based on the learner’s learning style and the relevant metadata of the content fragments, like
the pedagogical roles and media types.
4. Presentation layer: On this level authors specify how the learning resources should be
presented depending on the specific context, e.g. end user device. This can concern the
number of content fragments presented in parallel on a particular device.
6 Conclusion
This deliverable is aiming to map the current situation in the area of interoperability for adaptive learning
components. We have focused on semantic interoperability of user-adaptive systems, formal models and
standards, as well as access to metadata. We can state that in this field we are far from achieving
interoperability, since the different standards are not enough to realize it and therefore a mediation based
or Semantic Web based approach is still to be devised to reach something. This puts also the impressive
looking list of standards and tools in the field in a realistic perspective.
Interoperability in corporate settings has to include (Forte et al., 1999) also issues like distribution list
(which functions can receive which documents) related to various structures in different corporations, the
confidentiality level, and document classification schemes. In parallel with this document a deliverable on
privacy and data protection in corporate learning has been produced. Further we want to continue with
specification of personalized learning solutions at workplace and development of their prototypes
interfacing with corporate training systems.
7 References
Aguirre, S., Brantner, S., Huber, G., Markus, S., Miklos, Z., Mozo, A., Olmedilla, D., Salvachua, J., Simon,
B., Sobering, S., Zillinger, T. (2005). Corner Stones of Semantic Interoperability Demonstrated in a Smart
Space for Learning. In Proc. of the 2nd European Semantic Web Conference.
Brusilovsky, P. (2001). User Modeling and User-Adapted Interaction. Kluwer Academic Publishers, the
Netherlands, pp. 87-110.
Brusilovsky, P. (2003). Developing Adaptive Educational Hypermedia Systems: from Design Models to
Authoring Tools, Authoring Tools for Advanced Technology Learning Environments, Kluwer.
Page 14 of 16
PROLEARN Deliverable 1.2
CORDRA
(2004).
CORDRA:
Technical
Introduction
http://www.lsal.cmu.edu/lsal/expertise/projects/cordra/intro/intro-v1p00.html
and
Overview.
Dolog, P. (2004). Identifying relevant fragments of learner profile on the semantic web. In Proc. Of
SWEL’2004 — Intl. Workshop on Semantic Web for eLearning, Intl. Semantic Web Conference 2004,
Hiroshima, Japan, November 2004.
Dolog, P., & Schäfer, M. (2005). A framework for A Framework for Browsing, Manipulating and
Maintaining Interoperable Learner Profiles. In Proc. of UM’2005. Springer LNAI.
Forte, E., Haenni, F., Warkentyne, K., Duval, E., Cardinaels, K., Vervaet, E., Hendrikx, K., Wentland
Forte, M., Simillion, F. (1999). Semantic and Pedagogic Interoperability Mechanisms in the ARIADNE
Educational Repository. ACM SIGMOD Record, Vol. 28, No. 1, pp. 20-25.
Hatala, M., Richards, G., Eap, T., Willms, J. (2004). The Interoperability of Learning Object Repositories
and Services: Standards, Implementations and Lessons Learned. In Proc. of the 13th World Wide Web
Conference. New York City, USA, 2004.
Hodgins, W. (2005) Grand Challenges for Learning Objects. Learntec, Karlsruhe.
IMS (2003). IMS Digital Repositories Interoperability - Core Functions
http://www.imsglobal.org/digitalrepositories/driv1p0/imsdri_infov1p0.html
Information
Model.
IST (2004). Information Society Technologies priority. http://www.cordis.lu/ist/
Koper, R. (2001). Modelling units of study from a pedagogical perspective: the pedagogical metamodel
behind EML.
Koper, R., Olivier, B. (2004) Representing the Learning Design of Units of Learning. Educational
Technology & Society, 7 (3), 97-111.
Koper, R. (2005) An Introduction to Learning Design. In Koper, R. and Tattersall, C. eds. Learning
Design. A Handbook on Modelling and Delivering Networked Education and Training, Springer, pp. 3-22.
Kravcik, M. (2004) Specification of Adaptation Strategy by FOSP Method. AH 2004: Workshop
Proceedings, TU/e, 429-435.
Kravcik, M., Specht, M. (2004) Authoring Adaptive Courses – ALE Approach. Proc. of the Web Based
Education 2004 Conference. ACTA Press
Kravcik, M., Specht, M., Oppermann, R. (2004) Evaluation of WINDS Authoring Environment. Proc. of the
Adaptive Hypermedia 2004 Conference. Springer
LOM (2002). IEEE Standard for Learning Object Metadata. IEEE
Nejdl, W., Wolf, B., Qu, C., Decker, S., Sintek, M., Naeve, A., Nilsson, M., Palmer, M., and Risch, T.
(2002). EDUTELLA: a P2P Networking Infrastructure based on RDF. In Proc. of 11th World Wide Web
Conference, 604–615, Hawaii, USA, May 2002.
OKI (2004). Open Knowledge Initiative: http://web.mit.edu/oki/
OpenURL (2004). OpenURL, http://library.caltech.edu/openurl/
QEL (2004). Edutella Query Exchange Language http://edutella.jxta.org/spec/qel.html
RDF (2005). RDF, http://www.w3.org/RDF.
Simon, B., Massart, D., van Assche, F., Ternier, S., Duval, E., Brantner, S., Olmedilla, D., Miklos, Z.
(2005). A Simple Query Interface for Interoperable Learning Repositories. In Proc. of the WWW 2005
Conference.
Page 15 of 16
PROLEARN Deliverable 1.2
SPARQL (2005). SPARQL Overview, http://www.w3.org/2005/Talks/12May-SPARQL/all.html
SQI (2005). Simple Query Interface Specification. http://www.prolearn-project.org/lori
SRU (2005). Search and Retrieve URL Service. http://www.loc.gov/z3950/agency/zing/srw/sru.html
SRW (2004). ZING, Search/Retrieve Web Service (SRW): http://lcweb.loc.gov/z3950/agency/zing/srw/
Verbert, K., & Duval, E. (2004). Towards a global architecture for learning objects: a comparative analysis
of learning object content models. In Proc. of the ED-MEDIA 2004 World Conference on Educational
Multimedia, Hypermedia and Telecommunications, AACE, pp. 202–209.
ZING (2001). Z39.50, “Z39.50: Part 1 - An Overview,” Biblio Tech Review.
Page 16 of 16