Content uploaded by Peter Dolog
Author content
All content in this area was uploaded by Peter Dolog on Oct 26, 2012
Content may be subject to copyright.
Personalization in Distributed e-Learning Environments
Peter Dolog, Nicola Henze, Wolfgang Nejdl ∗, Michael Sintek†
ABSTRACT
Personalized support for learners becomes even more important,
when e-Learning takes place in open and dynamic learning and in-
formation networks. This paper shows how to realize personalized
learning support in distributed learning environments based on Se-
mantic Web technologies. Our approach fills the existing gap be-
tween current adaptive educational systems with well-established
personalization functionality, and open, dynamic learning reposi-
tory networks. We propose a service-based architecture for estab-
lishing personalized e-Learning, where personalization functional-
ity is provided by various web-services. A Personal Learning As-
sistant integrates personalization services and other supporting ser-
vices, and provides the personalized access to learning resources in
an e-Learning network.
Categories and Subject Descriptors
H.3.3 [Information storage and retrieval]: Information Search
and Retrieval—query formulation; H.3.4 [Systems and Software]:
[Distributed systems, Information networks, User profiles and alert
services]; H.3.5 [Online Information Services]: [Web-based ser-
vices]; H.5.4 [Information Interfaces and Presentation]: Hyper-
text/Hypermedia—Architectures, Navigation, User issues; K.3.1
[Computer Uses in Education]: [Distance learning]
General Terms
Standardization, Human Factors
Keywords
Personalization, Adaptation, P2P, Learning Repositories, Stan-
dards, Ontologies, Web Services
1. INTRODUCTION
Personalized learning using distributed information in dynamic
and heterogeneous learning environments is still an unsolved prob-
lem in e-Learning research. We envision a connected network
∗L3S Research Center, University of Hannover, Expo Plaza
1, D-30539 Hannover, Germany, {dolog, henze, nejdl}@
learninglab.de, http://www.learninglab.de/˜dolog,
http://www.kbs.uni-hannover.de/˜{henze,nejdl}
†German Research Center for Artificial Intelligence (DFKI)
GmbH, Knowledge Management Group, Postfach 2080,
D-67608 Kaiserslautern, Germany Michael.Sintek@dfki.de,
http://www.dfki.uni-kl.de/˜sintek
Copyright is held by the author/owner(s).
WWW2004, May 17–22, 2004, New York, New York, USA.
ACM 1-58113-912-8/04/0005.
of learning management and educational systems where learners
will be individually supported in accessing learning resources, tak-
ing part in courses or learning activities, entering communication
rooms, etc. In this setting, authors of learning materials will be in
full control over their content, learning resources, and courses.
Several approaches in this direction are currently investigated,
ranging from federated or distributed learning repositories (cf.
ARIADNE [4] or EDUTELLA [20]) or brokerage platforms (cf.
UNIVERSAL [21]), which focus on the dynamic and networking
aspects, learning management systems (cf. [27, 5]), which focus on
course delivery and administrative aspects, and adaptive web-based
educational systems (cf. [9, 22, 41]) which offer personalized ac-
cess and presentation facilities to learning resources for specific
application domains.
In the ELENA project (www.elena-project.org) we are
currently working on solutions to provide personalization, open-
ness, and interoperability [36] in the context of smart spaces for
learning. In particular, we investigate how to integrate the advan-
tages of open learning repositories with strategies and techniques
successfully employed in web-based educational systems, espe-
cially methods and techniques developed for adaptive educational
hypermedia systems.
Research in adaptive educational hypermedia has ascertained
several techniques for navigational level and content level adap-
tation (for an overview of terms and ideas of adaptive (educational)
hypermedia, we refer the reader to [10]), and has let to the hypoth-
esis that at least some techniques used in adaptive educational hy-
permedia can be encapsulated in separate adaptation modules [24].
There are several characteristics of open learning repositories,
integrating heterogeneous resource providers, which distinguish
them from most other currently studied systems. First of all, re-
sources can appear and disappear in ad-hoc manner. In addition,
peers providing resources can appear and disappear, too. Resources
are authored by different people with different goals, background,
domain expertise, etc. Providers of a resource can maintain the
resource in proprietary databases. They might already have some
personalization techniques implemented for the purposes of their
specific context. They might employ user or learner models (which
usually reflect applied techniques as well). User or learner features
can already be maintained in human resource management systems,
task management systems or user modeling servers. Furthermore,
resources are accessed and consumed by people which differ in a
wide range of characteristics.
Learning in open environments demands even more effective
personalization approaches to provide learner orientation and in-
dividualized access support [10].
The open problems in the context of personalization we are try-
ing to address are:
•How to provide personalization capabilities making use of
distributed yet connected repositories.
•How to support learner identification and profiles in such a
distributed environment.
•How to integrate personalization capabilities with other func-
tionalities needed to provide support for learners.
In this paper we describe an approach which provides personal-
ization capabilities based on distributed services. We employ se-
mantic web technologies to represent knowledge about resources,
learners, and services and investigate an architecture which inte-
grates distributed learning repositories and services without the
need of centralized control.
The paper is structured as follows: First, we motivate our work
by a simple scenario of learning in an open e-Learning network.
Our design of an adaptive semantic web infrastructure facilitated
by a service-based architecture is described in section 3. Section 4
shows how we use ontologies and metadata descriptions for various
types of resources in the e-Learning domain. Section 5 describes
the current state of implementation. After a comparison to related
work in section 6 the paper ends with conclusion and remarks on
further work.
2. PERSONALIZED LEARNING SUPPORT
IN DISTRIBUTED ENVIRONMENTS
In this section, we describe a simple e-Learning scenario from
the ELENA project to motivate our approach. Consider the fol-
lowing situation: A company starts a new software project. As the
user interface should be made available via the Internet, the com-
pany decides to implement the whole project in Java. The company
hires new people for that project, which need to be trained in Java
programming. Because the company does not have much experi-
ence and knowledge on Java programming, they decide to register
in an e-Learning network in order to search for appropriate learning
resources.
A member of the company, who already has programming expe-
rience in some other programming languages, wants to know how
specific programming concepts are realized in the Java program-
ming language. For example, she wants to know how to implement
concurrent programs in Java. This user will submit a query for
learning resources on “concurrent programming” and “Java” using
a personalized search service, which enriches the request with user
profile information (like information about her knowledge in pro-
gramming languages, her preferences for teaching language, style,
etc.). The user retrieves from the network learning resources in
her preferred language that teach “concurrent programming” in the
Java programming language. Learning resources, that are targeted
to experienced learners, are highlighted. We call this functionality
personalized search.
Retrieved learning resources are enriched with pointers to other,
related and relevant information. Links to relevant examples, dif-
ferent explanations, more detailed descriptions, etc., are provided.
In addition, the context of a learning resource, for example in a
course, can be provided for user. We call this functionality person-
alized link generation.
Investigating the scenario in more depth, we see that recommen-
dations for learning resources have to take several issues into ac-
count: First, suggested learning resources need to fit to language
constraints, device constraints, costs, etc. Second, they should fit
to the experience of the user who e.g. already has some knowledge
in other programming languages, etc. The context of a learning re-
source, e.g. the course or the courses where it has been introduced,
or related examples, exercises or projects referring to the specific
learning content of the resource, etc., can be analyzed.
To facilitate learning in our scenario, several functionalities need
to be provided. It is necessary to handle various types of meta-
data for resources in an open network, describing learners, learning
resources, information provided by the resources, as well as per-
sonalization strategies. We need to provide facilities for entering
a user query, and the translation of this to various formal query
languages supported in the network is required. Furthermore, the
query should reflect user preferences so we need to transform the
query to a new query which incorporates relevant user preferences.
Several technologies have been developed for shaping, con-
structing and developing the semantic web. RDF/S [30, 8] and its
extensions like DAML+OIL [14] and OWL [38] have been devel-
oped to define metadata schemas, domain ontologies and resource
descriptions. In the e-Learning domain there are standards emerg-
ing which describe learning resources, among them RDF-bindings
of LOM (Learning Objects Metadata [33] or Dublin Core [19].
Learners can be described using the IEEE PAPI (Public and Private
Information for Learners [26]) or the IMS LIP (Learner Informa-
tion Package Specification [28]) specifications.
The DAML-based Web service ontology (DAML-S [13]) is an
example of an initiative which supplies Web service providers with
a core set of markup language constructs for describing properties
and capabilities of their Web services in unambiguous, computer-
interpretable form. The aim of DAML-S markup of Web services
is to facilitate the automation of Web service tasks, including au-
tomated Web service discovery, execution, composition and inter-
operation. DAML-S provides a possibility to describe service pro-
files, process models, and bindings to an accessibility protocol and
ports through witch a particular service is available (e.g. the web
service description language WSDL [40] with its bindings to the
Simple Object Access Protocol [39] (SOAP), or GET/POST for
HTTP).
The TRIPLE rule, transformation and query language for the
semantic web has been introduced to reason over distributed an-
notations of resources. TRIPLE is able to handle the semantic
web descriptions formats like those previously mentioned (see ap-
pendix 10 for brief introduction).
3. SERVICES FOR PERSONALIZATION
ON THE SEMANTIC WEB
Our architecture for an adaptive educational semantic web ben-
efits from the following semantic web technologies: Information
and learning resources provided in various connected systems can
be described using OWL. Services which carry out personaliza-
tion functionality like personalized search or personalized recom-
mendations, as well as other required learning services, can be de-
scribed in DAML-S, and are accessible via WSDL and SOAP, the
functionalities identified in our e-Learning scenario can be encap-
sulated into services, possibly composed of other services. This re-
quires seamless integration and flow of results between the services
and seamless presentation of results to a user, as shown in fig. 1. In
the following, we will describe the services identified in this figure,
as well as some additional services important in the context of an
Adaptive Educational Semantic Web.
3.1 Personal Learning Assistant And User In-
teraction
Personal Learning Assistant Service.The central component of
our personalization service architecture is the Personal Learning
(Adaptive Educational) Semantic Web
Personal Learning Assistant
Services
User
Metadata
Link Generation
Services
Recommendation
Services
Annotation
Services
Mapping Services
Resource Provision Network (e.g. Edutella P2P
Infrastructure)
Resource
Metadata
Resource
Metadata
Resource
Metadata
User
Interaction
Component
Resource
provider
peer A Resource
provider
peer B
Resource
provider
peer C
Ontology Services
Learner
Repository Services
Query
Service
Query Rewriting
Services
Modification
Service
Figure 1: An architecture for personalization services
Assistant (PLA) Service which integrates and uses the other ser-
vices described in the following sections to find learning resources,
courses, or complete learning paths suitable for a user.
In future, the PLA Service will be able to search for suitable
service candidates, and to combine them (“service discovery and
composition”).
User Interaction Components.The PLA Service is either ex-
posed via an HTTP GET/POST binding, thus allowing direct in-
teraction with a user by means of a web browser, or is accessed
by separate User Interaction Components. To support learners with
different device preferences several types of these User Interaction
Components may be implemented: web-based, PDA-based, special
desktop clients, etc.
Our User Interaction Component provides a search interface in-
teracting with a subject ontology to construct appropriate queries,
as well as a user interface for refining user queries when they have
been constructed using subjects which do not match entries in the
particular subject ontology. The subject ontology service is able
to provide similar entries to the ones typed in the search interface.
Furthermore, the User Interaction Component visualizes the results
of a query, as well as additional personalization and annotation
hints.
3.2 Personalization Services
Query Rewriting Service.The Query Rewriting Service extends
a user query by additional restrictions, joins, and variables based
on various profiles. This extension is performed based on heuristic
rules/functions maintained by the Query Rewriting Service.
Query Rewriting Services can be asked for adding additional
constraints to user queries based on user preferences and language
capabilities. They can also be asked to extend a user query based
on previous learner performance maintained in learner profiles, if a
query is constructed in the context of improving skills.
Query Rewriting Services can also be asked to rewrite a user
query based on information the connected services need, which can
be exposed as input part in DAML-S based service profile descrip-
tions.
Recommendation Service.The Recommendation Service pro-
vides annotations for learning resources in accordance with the in-
formation in a learner’s profile. These annotations can refer to the
educational state of a learning resource, the processing state of a
learning resource, etc. The service holds heuristic rules for deriv-
ing recommendations based on learner profile information. Rec-
ommendation Services can be asked to add recommendation infor-
mation to existing instances based on learner profile information.
Link Generation Service.A Link Generation Service provides
(personalized) semantic relations for a learning resource in accor-
dance with the information in a learner’s profile. These relations
can show the context of a resource (e.g. a course in which this
learning resource is included), or they can show other learning re-
sources related to this resource (e.g., examples for this learning re-
source, alternative explanations, exercises). The Link Generation
Service holds heuristic rules for creating semantic hypertext links.
Some of the rules refer to information from the learner profile, in
absence of learner profile information the service can at least pro-
vide some, not optimized, hypertext links.
Link Generation Services can be asked for adding links and link
type annotations to a given learning resource. They can be asked
to generate a context for a given learning resource, or to generate
a context for several learning resources by adding hyperlinks be-
tween them.
3.3 Supporting Services
Ontology Service.An Ontology Service holds one or several on-
tologies and can be asked to return a whole ontology, a part of it
(e.g., a subgraph selected via some filter criterion), or can answer
queries of the kind “give me all subconcepts of concept C”, “which
properties are defined for concept C”, “who authored concept C”,
etc. Since ontologies will change over time, Ontology Services also
have to accept update requests and inform other services of these
updates.
Mapping Service.Mapping Services hold mappings between on-
tologies (or schemas) to allow services not using the same ontolo-
gies to communicate with each other. Such a Mapping Service can
be asked, e.g., to map a concept Cfrom one ontology to a concept
C0in another ontology, or to map an instance Iformulated in terms
of one ontology to an instance I0formulated in terms of another on-
tology. Since ontologies change over time, Mapping Services also
need to understand requests for updating the mapping specifica-
tions.
Repository Services.In general, Repository Services provide
access to any kind of repository which is connected to a net-
work. Repositories can be simple files, single databases, federated
databases, or a P2P network infrastructure.
A Repository Service maintains a link to a metadata store. This
might be a physical connection to a database or might be a group of
peers with an address (identification) of subnetworks where query
or manipulation commands will be submitted.
Repository Services can be of two kinds: Query Services and
Modification Services (for insert, update, or delete operations). The
repository provider can be asked to return references to resources
matching a given query, to create a new reference to a resource with
its new metadata, to delete a reference to a resource and its meta-
data, and to modify resource metadata. We assume that a Query
Service receives queries in its query language. These queries are
expressed using ontologies understood by the service, so the call-
ing service (e.g., the PLA) must provide the query in the correct
language (possibly using additional mapping/query transformation
services), or the storage service provider must contact other ser-
vices to get the appropriate format of a query.
Edutella services [32] are examples of such Repository Services
which access a P2P - Resource Provision Network. Edutella pro-
vides possibilities to connect repositories by implementing a so
called provision interface. Through this interface a learning repos-
itory can expose its metadata to the P2P network. Edutella also
provides a storage service to query the Edutella network by imple-
menting a consumer query interface. Edutella peers communicate
using a common internal data model. An RDF and Datalog based
query language QEL[34] is provided through the consumer query
interface together with a definition of the query result format. The
consumer interface provides the possibility to ask for a query or to
modify metadata stored in the network.
Further Services.Other services for authoring learning materials
and metadata / annotations for them, as well as services for learner
assessment might be useful as well. In addition to passive learn-
ing objects returned by PLA services, additional learning services
might provide educational activities to the users like distributed
classroom sessions and tutoring sessions.
4. METADATA AND ONTOLOGIES
As the scenario discussed in section 2 has shown, we need in-
formation about resources and participants involved in the learn-
ing situation, using appropriate standards wherever possible. These
standards specify properties for resources, and usually group them
into appropriate categories. [16] discusses the usefulness of such
standards for open e-learning environment in the context of per-
sonalization.
Concept
isPrerequisiteFor Instance* Concept isPrerequisiteFor*
LearningResource
dcterms:hasPart Instance* LearningResource
dc:subject Instance* Concept
dc:description String
dc:creator Instance lom:entity
dc:title String
...
dc:subject*
dcterms:hasPart* dcterms:requires*
LearningUnit
duration String
isa
Example
source Instance :THING
isa
Course
Qualification String
Location String
isa
Figure 2: An excerpt of ontology for learning resources
On the other hand these standards still have shortcomings (see
e.g. [2, 35]), the main one being their exclusive focus on property
based specifications. Semantic web technologies allow us to en-
hance these specifications using classes of objects with common
attributes. Another shortcoming of the standards is that they do
not include any domain ontologies - which again can be specified
building on semantic web formalisms. In the ELENA project we
therefore represent the properties specified in e-Learning standards
as properties of appropriate RDF classes. In addition, we employ
several domain ontologies which are either based on standardized
classification systems or which are specific for our courses.
Describing Learning Resources.An excerpt of a learn-
ing resource ontology is depicted in fig. 2. The class
LearningResource specifies common attributes used to de-
scribe resources in a subnetwork of the resource provision net-
work. The attributes are adopted from Dublin Core and Dublin
Core Terms standards [19]. The four subclasses depicted in fig. 2
refer to special kinds of learning resources. Course is a learn-
ing resource which can have a Location. It results in specific
Qualifications. A LearningUnit is a learning resource
with specific duration. Examples usually explain particular con-
text of a concept or subject being taught represented by source
(a particular project or specific situation).
LearningResource-s and their subclasses can be composite
structures (dcterms:hasPart relation). They can be also con-
nected through a prerequisite relation (dcterms:requires).
The class Concept is used to describe concepts as main informa-
tion entities from domain knowledge communicated by the learn-
ing resources. Concept and LearningResource are related
by the dc:subject property. Concepts can be chained through
an isPrerequisiteFor relation if prerequisites are defined on
the concept level and not on the learning resource level.
The structure of the learning resource metadata can be extended,
for example by a slot for annotation with the role of the learning re-
source, its type, level of covering or roles of particular concepts in
the learning resource. These additional relations can enhance adap-
tation possibilities for construction of learning sequences based on
user profile, annotating the position of a user in the learning re-
source structure, helping to identify main outcomes of a learning
resource based on roles and level of concept coverage, and so on.
OO_Class
Object_Oriented
isa
OO_Inheritance
isa
OO_Method
isa
OO_Object
isa
OO_Interface
isa
Logical
Programming_Strategies
isa isa
Imperative
isa
Functional
isa
Figure 3: An excerpt of concept ontology for Java e-lecture
Describing Domains.Specific domain information is usually
described by concepts and their mutual relationships in a do-
main. Domain concept models can form complex structures.
In our example we show just a fragment of a domain knowl-
edge base covering Java programming concepts, and include the
isa (subConceptOf) relationship between these concepts. Fig-
ure 3 depicts the Programming Strategies concept with its
subconcepts: Object Oriented,Imperative,Logical,
and Functional.OO Class,OO Method,OO Object,
OO Inheritance, and OO Interface are depicted as subcon-
cepts of Object Oriented. Other relations between concepts
might be useful for personalization purposes as well, e.g. sequenc-
ing or dependency relations.
Learner
hasPreference Instance* lip_rdfs:Preference
hasSecurityAndPrivacy Instance* privacy_rdfs:SecurityAndPrivacy
hasPortfolio Instance* papi_rdfs:Portfolio
hasPerformance Instance* papi_rdfs:Performance
learnerId String
...
papi_rdfs:Portfolio
papi_rdfs:PortfolioType String
papi_rdfs:PortfolioCertificate Instance papi_rdfs:Certificate
papi_rdfs:PortfolioSource String
papi_rdfs:PortfolioId String
papi_rdfs:portfolio_privacy Instance privacy_rdfs:PrivacyInfo
hasPortfolio*
papi_rdfs:Performance
papi_rdfs:Valid_from String
papi_rdfs:learning_competency Instance* Competencies_rd:RDCEO
papi_rdfs:Concept
papi_rdfs:Issued_date String
papi_rdfs:PerformanceId String
papi_rdfs:PerformancePortfolio Instance* papi_rdfs:Portfolio
...
hasPerformance*
lip_rdfs:Preference
lip_rdfs:hasImportanceOver Instance lip_rdfs:Preference
lip_rdfs:preferencePrivacy Instance privacy_rdfs:PrivacyInfo
lip_rdfs:ValidTo String
lip_rdfs:ValidFrom String
hasPreference*
lip_rdfs:hasImportanceOver
papi_rdfs:PerformancePortfolio*
Figure 4: An excerpt of learner ontology
Describing Learners.Information about learners is needed to be
able to recommend appropriate learning resources which are rel-
evant with respect to user interests, user performance in different
courses within one domain or even different domains, user goals
and preferences, and so on.
Figure 4 shows a subset of a learner ontology (for more complex
user models see, e.g., [18]). It includes the class Learner, related
to other classes like performance, preference, and portfolio. The
portfolio is for maintaining learning resources, which have been
created or accessed during learning. This learner model was created
by enhancing core parts of IEEE PAPI and IMS LIP with some
specific extensions for the ELENA project.
5. IMPLEMENTING THE SERVICES
Based on our design described in section 3, we implemented a
first software prototype, which we will describe in the following
section. Figure 5 depicts the UML collaboration diagram showing
a message flow between service providers we have implemented
for the ELENA PLA. Boxes represent service providers, lines rep-
resent links (dependencies) between the providers. A direction of
a message or invoking operation is indicated by a small arrow on
top of a line with the name and parameters of that operation. We
use two kinds of arrows in fig. 5. The normal arrow (→) is used to
indicate a plain message. The “harpoon” (+) indicates explicitly
that a message is asynchronous. Square brackets are used to indi-
cate a condition which enables a certain message to be passed: If
the condition is not satisfied the message is not sent.
The PersonalizedSearchService provides a user in-
terface for searching and displaying personalized results to a
user. A user can send two messages through the provided
user interface. First the message (userQuery) notifies the
PersonalizedSearchService about user, text typed in
fields or concepts selected from the ACM classification hier-
archy, and whether to provide personalization information or
not. If the user typed a free text into fields provided, the
PersonalizedSearchService contacts an ontology service
(in our case the ACMOntologyService) to get concepts simi-
lar to the text typed (the message getSimilarConcepts). The
PersonalizedSearchService then displays these concepts
to a user to refine his/her query. After selecting precise concepts
from suggested entries from the ontology, the user can send a re-
fined request to the PersonalizedSearchService.
The PersonalizedSearchService notifies the
PLAService about the user query (the query message).
The PLAService first makes use of the MappingService
provider to generate a QEL query by sending the generateQEL
message. The service constructs an appropriate QEL query
from the concepts list. In addition, the PLAService contacts
the QueryRewritingService provider after receiving the
QELQuery to rewrite the QELQuery according to a learner
profile, adding additional constraints to the QELQuery.
PLAService sends a message with the rewritten QELQuery
to a QueryService, in our case the Edutella query service which
propagates the query into the Edutella P2P resource provision net-
work. The Edutella QueryService returns all query results.
If the learner prefers recommendation information in-
cluded with the query results, the PLAService contacts the
RecommendationService to derive such recommenda-
tion information according to the learner profile or to group
profiles (collaborative recommendation). When such person-
alized results are available, the PLAService notifies the
PersonalizedSearchService to display the results to a
learner.
5.1 Personal Learning Assistant Services
The Personal Learning Assistant Service (PLA) aims at con-
necting and integrating the services which are needed to perform
the learning support task. Personalized Search for example con-
nects mapping, query rewriting, query, and recommendation ser-
vices. We are working on providing other learning support services
like learning path generation service, course delivery services and
booking services.
PersonalizedSearchService
1: userQuery(user, list, personalization)
1.3: refinedQuery(user, conceptList)
PLAService
1.2 [Free text typed] displayConcepts(user,conceptList)
8: displayResults(QELResults)
2. query(user, conceptList)
ACMOntologyService
1.1 [Free text typed] conceptList:=getSimilarConcepts(List)
QueryRewritingService
QueryService MappingService
RecommendationService
4: QELQuery:=rewriteQEL(user, query)
3: QELQuery:=generateQEL(conceptList)
7.1: [personalization] LOMMetadata:=transformToLOM(QELResults)
7.3: [personalization] QELResults:=transformToQELResults(LOMMetadata)
5: sendQuery(QELquery)
7.2: [personalization] addRecommendation(user, LOMMetadata)
6: sendResults(QELquery)
7: [personalization] personalizeResults(user, QELResults)
Figure 5: A collaboration diagram of current implementation.
Figure 6: A prototype for search user interface.
Visualization.Figure 6 depicts a user interface for formulating a
user query for a particular concept or competence a user would
like to acquire, combined with a user interface providing results
with recommendation information represented by the traffic light
metaphor. Using this metaphor, a green ball marks recommended
learning resources, a red ball marks non-recommended learning re-
sources and a yellow ball marks partially recommended learning
resources.
The user interface is generated by a service which uses the cho-
sen ontology service (the ACM ontology service). List of learners
who have a learner profile maintained at the PLA service chosen is
displayed as well.
Users can type free text into three provided fields or can select
concepts from an ontology provided (in our example figure the user
typed “intelli agent”).
The user interface returning the results is generated according
to the concepts chosen and includes the query results returned by
the query service and personalized by the recommendation ser-
vices chosen at the PLA service. The personal recommendation
is depicted in the first column (PReco). There is a second column
(Reco), which provides learners with a group-based recommenda-
tion. The group-based recommendation is calculated according to
recommendations of learners from the same group.
We are working on further improvements of our prototype user
interfaces. This includes a user interface for specifying more com-
plex queries and a result interface pointing to further information
or directly to services for booking and delivery of learning services
and resources.
5.2 Personalization Services
Query Rewriting Service.We have implemented a query rewrit-
ing service which adds additional constraints to a QEL query cre-
ated according to which concepts a user selected. These constraints
reflect concepts and language preferences maintained in user pro-
files.
We illustrate the query rewriting principle on the following sim-
ple restriction profile, implemented in TRIPLE.
@edu:p1 {
edu:add1[rdf:type -> edu:AddSimpleRestriction;
rdf:predicate -> dc:lang;
rdf:object -> lang:de].
edu:add2[rdf:type -> edu:AddTopicRestriction;
edu:addTopic -> acmccs:’D.1.5’].}
This heuristic is used to extend a QEL query with a constraint
which restricts the results to learning resources in German language
(restriction edu:add1).
Another restriction derived from the user profile is a restriction
on resources about object-oriented programming (edu:add2).
The ACM Computer Classification System [1] is used to encode the
mentioned subject. In that classification system, the object-oriented
programming can be found in the category Drepresenting software.
The subcategory D.1 represents programming techniques with the
fifth subcategory being object-oriented programming. Heuristics
for query rewriting especially in case of concept or subject restric-
tions are usually more complex. They depend on concepts being
selected or typed as a user query.
The derived restrictions profile is used in a TRIPLE view which
takes as an input the profile and QEL query model. One of the rules
for reasoning over language restrictions profiles follows. The view
@edu:p1 encapsulates the restrictions model.
FORALL QUERY, VAR, PRED, OBJ, NEWLIT
QUERY[edu:hasQueryLiteral -> edu:NEWLIT] AND
edu:NEWLIT[rdf:type -> edu:RDFReifiedStatement;
rdf:subject -> VAR;
rdf:predicate -> PRED;
rdf:object -> OBJ]
<-
EXISTS LITERAL, ANY (
QUERY[rdf:type -> edu:QEL3Query;
edu:hasQueryLiteral -> LITERAL]
AND
LITERAL[rdf:type -> edu:RDFReifiedStatement;
rdf:subject ->
VAR[rdf:type -> edu:Variable];
rdf:predicate -> dc:ANY])
AND
EXISTS A
A[rdf:type -> edu:AddSimpleRestriction;
rdf:predicate -> PRED;
rdf:object -> OBJ]@edu:p1
AND
unify(NEWLIT, lit(VAR,PRED,OBJ)).
Recommendation Service.The recommendation service pro-
vides the following functionality: It can annotate learning resources
according to their educational state for a user. E.g. it can recom-
mend a resource to a specific user, or give a less strong recommen-
dation like might be understandable. Furthermore, it can not rec-
ommend a learning resource or point out that this learning resource
leads to a page that the user has already visited.
To derive appropriate recommendation annotations for a partic-
ular user, prerequisite concepts for a learning resource have to be
mastered by the user. The lr:isPrerequisiteFor relation-
ships of concepts covered in a learning resource are analyzed for
this purpose. On the other hand, a user performance profile and
competencies acquired and maintained in that profile are analyzed
in comparison to the prerequisites of particular learning resource.
One example of a recommendation rule is a rule which deter-
mines learning resources which are Recommended. A learning
resource is recommended if all prerequisite concepts of all of con-
cepts it covers have been mastered by a user:
FORALL LR,U learning_state(LR, U, Recommended) <-
learning_resource(LR) AND user(U)
AND NOT learning_state(LR, U, Already_visited)
AND FORALL Ck ( prerequisite_concepts(LR, Ck) ->
p_obs(Ck, U, Learned) ).
Predicates used in the rule derive concepts like learning resource,
concepts, and users, observations and learning states from metadata
based on types taken from ontologies described in section 4.
We have implemented other rules to compute less strong recom-
mendations. This includes for example a recommendation that a
resource Might be understandable if at least one prerequi-
site concept has been learned.
This kind of recommendation can be used for example as a link
annotation technique in the area of adaptive hypermedia [10], or to
annotate query results with the recommendation information. On
the user interface side, it is often implemented using the already
mentioned traffic lights.
Link Generation Service.A Link Generation Service connects a
learning resource to other learning resources, or it connects a learn-
ing resource to a context, e.g. within a course with links to previous
and next steps. As an example of Link Generation Service, we have
implemented a service that relates a learning resource to other re-
sources which provide related examples of the learning resource’s
content.
One example for deriving such an example-relation for a re-
source Ris by ensuring that each concept on Ris covered by the
example E:
FORALL R, E example(R,E) <-
LearningResource(R) AND example(E) AND
EXISTS C1 (R[dc:subject->C1]) AND
FORALL C2 (R[dc:subject->C2]->E[dc:subject->C2]).
The second line in the rule above ensures that Ris a
LearningResource and Eis an Example (using the ontol-
ogy for learning resources described in the section 4). The third
rule verifies that Rreally is about some concept - i.e. there exists a
metadata annotation like dc:subject. The fourth line then ex-
presses what our rule should check: Whether each concept on R
will be explained in the example E.
A user profile can be taken into account when generating the ex-
ample relationship. A personalized pedagogical recommendation
of an example might include an example showing new things to
learn in a context of already known / learned concepts: This would
embed the concepts to learn in the previous learning experience of
a user. The rule to derive this best example follows.
FORALL R, E, U best_example(R,E,U) <-
LearningResource(R) AND example(E) AND user(U)
AND example(R,E) AND FORALL C (
(E[dc:subject->C] AND NOT R[dc:subject->C])->
p_obs(C, U, Learned) ).
Further rules for generating personalized hypertext associations
can be implemented. Other relationships, classes and properties
from the domain, user, and learning resource ontology can be used
for these purposes [17]. The isa relationship in the concept-
ontology of the java application domain can be utilized to rec-
ommend learning resources either more general, e.g. introducing
a concept of programming strategies, or more specific concepts.
The sequencing relationship can be used to recommend learning
resources in the following way: A resource which describes a con-
cept (the concept appears in the dc:subject property for the re-
source) from the beginning of the sequence will be recommended
earlier than a resource which describes a concept from the end of
such a sequence. A dependency relationship referring to whether
a concept depends on another concept can be used as well to rec-
ommend learning resources which describe dependent concepts to-
gether with a learning resource describing a concept which was
recommended by another rule.
5.3 Supporting Services
Query Service.The Edutella P2P infrastructure [32] allows us
to connect peers which provide RDF metadata about resources.
Edutella also provides us with a powerful Datalog-based query lan-
guage, RDF-QEL. A query can be formulated in RDF format as
well, and it can reference several schemas. An example for a sim-
ple query over resources is the following:
s(X, <dc:title>, Y),
s(X, <dc:subject>, S),
qel:equals(S, <java:OO_Class>).
The query tries to find resources where dc:subject equals
to java:OO Class. The prefixes qel:,dc:, and java: are
abbreviations for URIs of the schemas used. Variable Xwill be
bound to URIs of resources, variable Ywill be bound to titles of the
resources, and variable Swill be bound to subjects of the resources.
QEL offers a full range of predicates besides equality, general
Datalog rules, and outer join (see [34]). Not all predicates need to
be supported by peer providers. The QueryService exposes an
interface to Edutella for querying. A client of that service can send
a message containing a QEL query to that service.
Mapping Service.We have implemented a mapping service for
mapping QEL variable bindings to LOM RDF bindings and back.
This was needed because our recommendation service accepts in-
put in LOM RDF bindings. On the other hand, additional recom-
mendation information plus LOM metadata have to be transformed
back to QEL variable bindings because the personalized search ser-
vice uses QEL variable bindings as a result set. These transforma-
tions are again done in TRIPLE.
Concept mappings between different subject ontologies, differ-
ent ontologies for describing learners, and different learning re-
source ontologies are important as well. The TRIPLE view/model
mechanism allow us to specify and implement models which em-
bed rules for mappings between that ontologies [31]. Currently we
are implementing such mapping heuristics between the ontologies
used in different systems connected in the ELENA network.
ACM Ontology Service.We have implemented a simple version
of an ontology service for the ACM classification system and its
RDF LOM bindings. The current version of our ontology service
supports requests for getting the whole ontology using the HTTP
protocol as well as requests for getting “similar” concepts from the
ontology to the submitted text string.
6. RELATED WORK
Our approach is based on adaptive hypermedia research. Adap-
tive hypermedia has been studied for closed environments, i.e. the
underlying document space / the hypermedia system is known to
the authors of the adaptive hypermedia system at design time of the
system. As a consequence, changes to this document space usually
cannot be considered: A change to the document space requires
the re-organization of the document space or at least some of the
documents in the document space.
First steps towards open adaptive e-Learning solutions have been
investigated in [23, 10, 16]. In this paper we extend this work by
moving towards even more decentralized solutions where both re-
sources and computation can be distributed. Besides personaliza-
tion services we introduce supporting services which are important
to realize the whole functionality of an adaptive educational seman-
tic web.
Similar to our approach, [12] builds on separating learning re-
sources from sequencing logic and additional models for adaptiv-
ity: Adaptivity blocks in the metadata of learning objects and in
the various models providing adaptivity like the narrative model,
candidate groups, etc. define the kind of adaptivity realizable with
the current piece of learning content. Driving force in these models
are the candidate groups that define how to teach a certain learning
concept. A rule engine selects the best candidates for each spe-
cific user in a given context. A shortcoming of the approach is
that the adaptivity requirements are considered only in the adaptiv-
ity blocks, while our approach considers all metadata as useful for
adaptation.
An early approach for defining an architecture for personaliza-
tion and adaptivity in the semantic web has been proposed in [3].
This approach is characterized by the transfer of ownership of
semantic web resources to the user, and therefore on the client
side. Versioning and other ownership-transfer related issues are
discussed. The authors motivate their approach by e-Business ap-
plications, and in particular by e-Procurement applications. The
domain of e-Learning has different requirements: not the optimiza-
tion of process-embedded tasks or repetition of tasks is relevant,
instead we want to provide guidance to novices in a complex infor-
mation space, point out relevant learning goals, learning materials,
or learning steps to take.
Personalized learning-portals are investigated in [11]. The learn-
ing portals provide views on learning activities which are provided
by so-called Activity Servers. The activity servers store both learn-
ing content and the learning activities possible with this special
content. A central student model server collects the data about stu-
dent performance from each activity server the student is working
on, as well as from every portal the student is registered to.
Comparing our work with standard models for adaptive hyper-
media systems such as the one used in AHA! [7], we observe that
they define several models like conceptual, navigational, adapta-
tional, teacher and learner models. Compared to our approach,
these models either correspond to ontologies / taxonomies, to
different schemas describing teacher and learner profile, and to
schemas describing the navigational structure of a course. We ex-
press adaptation functionalities as encapsulated and reusable Triple
rules, while the adaptation model in AHA uses a rule based lan-
guage encoded into XML. At the level of concept or information
items AHA! provides functionalities to describe requirements [6]
for the resource, which state what is required from a user to visit
that information.
7. CONCLUSION AND FURTHER WORK
In this paper we have described an approach to bring personal-
ization to the semantic web for the area of education and learning.
We have shown how personalization functionalities can be embed-
ded into semantic web services, supported by other services for re-
trieving learning resources or user information. We have discussed
our ELENA prototype implementing such services, connecting and
integrating them in our personal learning assistant.
Further research questions have to be investigated in the future.
One important issue in the semantic web context is the availabil-
ity of metadata as formal descriptions about information sources
whose quality has to be high enough to use them for sophisticated
services such as the ones discussed in this paper. Tools to sup-
port creation, maintenance and consistency between information
sources and metadata describing them have to be provided. Further
experiments with additional personalization methods derived from
the adaptive hypermedia system context and their critical evalua-
tion against the requirements of an open environment have to be
performed. Last but not least, we will investigate dynamic service
discovery and composition to support the reuse of personalization
functionalities in different contexts.
8. ACKNOWLEDGMENTS
This work is partially supported by FP5 EU/IST ELENA project
IST-2001-37264 (http://www.elena-project.org). We
would like to thank all members of the ELENA project consortium,
special thanks go to Bernd Simon, Barbara Kieslinger, and Tomaˇ
z
Klobuˇ
car.
9. REFERENCES
[1] The ACM Computer Classification System.
http://www.acm.org/class/1998/, 2002.
[2] ALLERT, H., RICHTER, C., AND NEJDL, W. Extending the
scope of the current discussion on metadata towards situated
models. In 5th International CSCL Conference (Bergen,
Norway, 2003).
[3] ARAGAO, V., FERNANDES, A., AND C.A.GOBLE. Towards
an architecture for personalization and adaptivity in the
semantic web. In Proc. 3rd. Intl. Conf. on Information
Integration and Web-Based Applications and Service, iiWAS
2001 (Linz, Austria, Sept. 2001).
[4] Ariadne: Alliance of remote instructional authoring and
distributions networks for europe, 2001.
http://ariadne.unil.ch/.
[5] BLACKBOARD INC. Blackboard course management
system. Blackboard INC:
http://www.blackboard.com/. Accessed: 20th
September 2003.
[6] BRA, P. D., AERTS, A., SMITS, D., AND STASH, N. AHA!
version 2.0. In Proceedings of the AACE ELearn’2002
conference (Oct. 2002), pp. 240–246.
[7] BRA, P. D., HOUBEN, G.-J., AND WU, H. AHAM: A
dexter-based reference model for adaptive hypermedia. In
Proc. of ACM Conference on Hypertext and Hypermedia
(Darmstadt, Germany, Feb. 1999), K. Tochtermann,
J. Westbomke, U. Wiil, and J. Leggett, Eds., pp. 147–156.
[8] BRICKLEY, D., AND GUHA, R. V. Resource Description
Framework (RDF) Schema Specification 1.0, 2002.
http://www.w3.org/TR/rdf-schema.
[9] BRUSILOVSKY, P., EKLUND, J., AND SCHWARZ, E.
Web-based Educations for All: A Tool for Development
Adaptive Courseware. In Proceedings of the Sevenths
International World Wide Web Conference, WWW’98 (1998).
[10] BRUSILOVSKY, P., AND NEJDL, W. Adaptive hypermedia
and adaptive web. In Practical Handbook of Internet
Computing, M. Singh, Ed. CRC Press, 2004. To appear.
[11] BRUSILOVSKY, P., AND NIJHAWAN, H. A framework for
adaptive e-learning based on distributed re-usable learning
activities. In In: M. Driscoll and T. C. Reeves (eds.)
Proceedings of World Conference on E-Learning, E-Learn
2002 (Montreal, Canada, 2002).
[12] DAGGER, D., WADE, V., AND CONLAN, O. Towards
“anytime, anywhere“ learning: The role and realization of
dynamic terminal personalization in adaptive elearning. In
Ed-Media 2003, World Conference on Educational
Multimedia, Hypermedia and Telecommunications (Hawaii,
2003).
[13] DAML-S SERVICES COALITION. DAML-S: Semantic
Markup for Web Services. Technical Report. Available at:
http://www.daml.org/services/daml-s/0.9/daml-s.pdf, May
2003.
[14] DAML+OIL, 2001.
http://www.daml.org/2001/03/daml+oil-index.html.
[15] DECKER, S., BRICKLEY, D., SAARELA, J., AND ANGELE,
J. A query and inference service for RDF. In QL’98 — The
Query Languages Workshop (Boston, USA, 1998),
WorldWideWeb Consortium (W3C).
[16] DOLOG, P., GAVRILOAIE, R., NEJDL, W., AND BRASE, J.
Integrating adaptive hypermedia techniques and open
rdf-based environments. In Proc. of 12th International World
Wide Web Conference (Budapest, Hungary, May 2003).
[17] DOLOG, P., HENZE, N., AND NEJDL, W. Logic-based open
hypermedia for the semantic web. In Proc. of International
Workshop on Hypermedia and the Semantic Web, Hypertext
2003 Conference, Nottinngham, UK (Aug. 2003).
[18] DOLOG, P., AND NEJDL, W. Challenges and benefits of the
semantic web for user modelling. In In Proc. of AH2003 —
Workshop on Adaptive Hypermedia and Adaptive Web-Based
Systems at World Wide Web, UserModelling and Hypertext
Conference 2003 (Budapest, Pittsburgh, Nottinngham, 2003).
[19] The Dublin Core Metadata Initiative.
http://dublincore.org/.
[20] Edutella, 2001. http://edutella.jxta.org/.
[21] GUTH, S., NEUMANN, G., AND SIMON, B. Universal -
design spaces of learning media. In Proceedings of the 34th
Hawaii International Conference on System Sciences (Maui,
Hawaii (USA), January 2001).
[22] HENZE, N., AND NEJDL, W. Adaptation in open corpus
hypermedia. IJAIED Special Issue on Adaptive and
Intelligent Web-Based Systems 12 (2001).
[23] HENZE, N., AND NEJDL, W. Knowledge modeling for open
adaptive hypermedia. In Proccedings of the 2nd International
Conference on Adaptive Hypermedia and Adaptive
Web-Based Systems (AH 2002) (Malaga, Spain, 2002).
[24] HENZE, N., AND NEJDL, W. Logically characterizing
adaptive educational hypermedia systems. In International
Workshop on Adaptive Hypermedia and Adaptive Web-based
Systems (AH 2003) (Budapest, Hungary, 2003).
[25] HORROCKS, I. The FaCT System, 2001. URL:
http://www.cs.man.ac.uk/∼horrocks/FaCT/.
[26] IEEE. IEEE P1484.2/D7, 2000-11-28. Draft Standard for
Learning Technology. Public and Private Information (PAPI)
for Learners (PAPI Learner). Available at:
http://ltsc.ieee.org/wg2/. Accessed on October 25, 2002.
[27] IMC INFORMATION MULTIMEDIA COMMUNICATION AG.
Clix - learning management system. IMC AG:
http://www.im-c.de/homepage/index.htm.
Accessed: 20th September 2003.
[28] IMS. IMS learner information package specification.
Available at: http://www.imsproject.org/profiles/index.cfm.
Accessed on October 25, 2002.
[29] KIFER, M., LAUSEN, G., AND WU, J. Logical foundations
of object-oriented and frame-based languages. Journal of the
ACM 42 (July 1995), 741–843.
[30] LASSILA, O., AND SWICK, R. W3C Resource Description
framework (RDF) Model and Syntax Specification. Available
at: http://www.w3.org/TR/REC-rdfsyntax/. Accessed on
October 25, 2002.
[31] MI KL ´
OS, Z., NEUMANN, G., ZDUN, U., AND SINTEK, M.
Querying semantic web resources using TRIPLE views. In
Proceedings of the 2nd International Semantic Web
Conference (ISWC2003) (Sanibel Island, Florida, USA, Oct.
2003).
[32] NEJDL, W., WOLF, B., QU, C., DECKER, S., SINTEK, M.,
NAEVE, A., NILSSON, M., PALM´
ER, M., AND RISCH, T.
EDUTELLA: a P2P Networking Infrastructure based on
RDF. In In Proc. of 11th World Wide Web Conference
(Hawaii, USA, May 2002).
[33] NILSSON, M. IMS Metadata RDF binding guide.
http://kmr.nada.kth.se/el/ims/metadata.html, May 2001.
[34] NILSSON, M., AND SIBERSKI, W. RDF Query Exchange
Language (QEL) - Concepts, Semantics and RDF Syntax.
Available at:
http://edutella.jxta.org/spec/qel.html.
Accessed: 20th September 2003, 2003.
[35] POLSANI, P. R. Use and abuse of reusable learning objects.
Journal of Digital information 3 (2003).
[36] SIMON, B., MIKLOS, Z., NEJDL, W., SINTEK, M., AND
SALVACHUA, J. Smart space for learning: A mediation
infrastructure for learning services. In Proc. of 12th
International World Wide Web Conference (Budapest,
Hungary, May 2003).
[37] SINTEK, M., AND DECKER, S. TRIPLE—A query,
inference, and transformation language for the semantic web.
In 1st International Semantic Web Conference (ISWC2002)
(June 2002).
[38] W3C. Owl web ontology language semantics and abstract
syntax. Technical Report. Available at:
http://www.w3.org/TR/owl-semantics/. Accessed: 20th
September 2003, Aug. 2003.
[39] W3C. Soap version 1.2 part 0: Primer. Technical Report.
Available at: http://www.w3.org/TR/soap12-part0/, June
2003.
[40] W3C. Web Services Description Language (WSDL) Version
2.0 Part 1: Core Language. Technical Report. Available at:
http://www.w3.org/TR/wsdl20/, Nov. 2003.
[41] WEBER, G., KUHL, H.-C., AND WEIBELZAHL, S.
Developing adaptive internet based courses with the
authoring system NetCoach. In Proceedings of the Third
Workshop on Adaptive Hypermedia, AH2001 (2001).
10. APPENDIX
TRIPLE [37] is a rule language for the Semantic Web which is
based on Horn logic and borrows many basic features from F-Logic
[29] but is especially designed for querying and transforming RDF
models. TRIPLE can be viewed as a successor of SiLRI (Simple
Logic-based RDF Interpreter [15]). One of the most important dif-
ferences to F-Logic and SiLRI is that TRIPLE does not have fixed
semantics for object-oriented features like classes and inheritance.
Description logics extensions of RDF (Schema) like OIL,
DAML+OIL, and OWL that cannot be fully handled by Horn
logic are provided as modules that interact with a description logic
classifier, e.g. FaCT [25], resulting in a hybrid rule language.
Namespaces and Resources TRIPLE has special support for
namespaces and resource identifiers. Namespaces are declared
via clause-like constructs of the form nsabbrev := namespace.,
e.g., rdf := ”http://www.w3.org/1999/02/22-rdf-syntax-ns#”.
Resources are written as nsabbrev:name, where nsabbrev is
a namespace abbreviation and name is the local name of the
resource.
Statements and Molecules Inspired by F-Logic ob-
ject syntax, an RDF statement (triple) is written as:
subject[predicate →object]. Several statements with
the same subject can be abbreviated as “molecules”:
edu:add1[rdf:predicate →dc:lang;rdf:object →lang:de;...].
Models RDF models, i.e., sets of statements, are made explicit
in TRIPLE (“first class citizens”).1Statements, molecules, and
also Horn atoms that are true in a specific model are written
as atom@model (similar to Flora-2 module syntax), where
atom is a statement, molecule, or Horn atom and model is a
model specification (i.e., a resource denoting a model), e.g.:
A[rdf:type →edu:AddSimpleRestriction]@edu:p1. TRIPLE also
1Note that the notion of model in RDF does not coincide with its
use in (mathematical) logics.
allows Skolem functions as model specifications. Skolem functions
can be used to transform one model (or several models) into a new
one when used in rules (e.g., for ontology mapping/integration):
O[P→Q]@sf(m1, X, Y )←− ....
Logical Formulae TRIPLE uses the usual set of connectives
and quantifiers for building formulae from statements/molecules
and Horn atoms, i.e., ∧,∨,¬,∀,∃, etc.2All variables must be
introduced via quantifiers, therefore marking them is not necessary
(i.e., TRIPLE does not require variables to start with an uppercase
letter as in Prolog).
Clauses and Blocks A TRIPLE clause is either a fact or a
rule. Rule heads may only contain conjunctions of molecules
and Horn atoms and must not contain (explicitly or implicitly)
any disjunctive or negated expressions. To assert that a set
of clauses is true in a specific model, a model block is used:
@model {clauses}, or, in case the model specification is param-
eterized: ∀Mdl @model(Mdl){clauses}.
2For TRIPLE programs in plain ASCII syntax, the symbols AND,
OR, NOT, FORALL, EXISTS, <-,->, etc. are used.
