E-Learning and Microformats: A Learning Object
Harvesting Model and a Sample Application
Ahmet Soylu1, Selahattin Kuru2, Fridolin Wild3, Felix Mödritscher3,
1
K. U. Leuven, Interdisciplinary Research on Technology, Education and Communication
(iTec), Campus Kortrijk,
Etienne Sabbelaan 53, 8500 Kortrijk, Belgium
ahmet.soylu@kuleuven-kortrijk.be
2
Işık University, Department of Computer Science and Engineering, Kumbaba Mevkii
Şile, 34980, Istanbul, Turkey
kuru@isikun.edu.tr
3
Vienna University of Economics and Business Administration,
Department of Information Systems, Augasse 2-6, 1090 Wien, Austria
{fridolin.wild,felix.moedritscher}@wu-wien.ac.at
Abstract. In order to support interoperability of learning tools and reusability
of resources, this paper introduces a framework for harvesting learning objects
from web-based content. Therefore, commonly-known web technologies are
examined with respect to their suitability for harvesting embedded meta-data.
Then, a lightweight application profile and a microformat for learning objects
are proposed based on well-known learning object metadata standards.
Additionally, we describe a web service which utilizes XSL transformation
(GRDDL) to extract learning objects from different web pages, and provide a
SQI target as a retrieval facility using a more complex query language called
SPARQL. Finally, we outline the applicability of our framework on the basis of
a search client employing the new SQI service for searching and retrieving
learning objects.
Keywords: E-Learning, Learning Object, Metadata, Semantic
Interoperability, SQI, RDF, SPARQL, GRDDL, Microformats.
Web,
1 Motivation
Learners are not bound neither to individual learning environments as closed box of
pure information nor to classical in-class learning environments anymore. Instead
they are facing with several tools (web 2.0) within their learning environment, which
enable them to collaborate, to reach an endless amount of information on the web, to
remix and share it, and to create social networks. The e-learning market has already
been overpopulated with tools and platforms to support different types of learning
communities with learning management, content management, and communication
facilities [1]. Depending on case, these tools are being used individually by learners,
or by means of mash-ups, or as heterogeneous systems which involve several tools
and might be centered around one particular learning system. All these aspects end up
in several technological challenges. Our work specially targets the following ones;
- Re-use: Learning content can be easily split up, re-used, and arranged again
in order to create different contexts.
- Interoperability: Multiple learning systems or tools can communicate with
each other to share and mix their learning resources.
- Accessibility: Learning content can be accessed by several learning tools and
devices, besides appropriateness of content can be ensured.
- Durability: Users are not bound to any specific system, tool or device,
interoperability and reusability is ensured.
We consider this work to be particularly important for mash-ups because the shift
towards learner centric approaches accompanied by an ever-increasing degree of
interaction and rising volumina of content provided by learners through the use of
diverse social and web 2.0 tools emphasizes the importance of collective and
personalized use of these distributed applications. Therefore, our work is considered
as an approach to address the re-usability, interoperability, accessibility, and
durability challenges of mash-ups by making existence, retrieval, sharing, and
harvesting of learning objects independent of their applications, where these
challenges were neglected due to the limiting features provided (RSS, import/export
facilities, etc.).
2 Semantic Web
Computers are not able to automatically apply actions like ‘derive’, ‘associate’, or
‘link’ onto information available through the web – in the way humans do. Instead,
they in general deal only with presentation and routine processing, as the web does
most often does not provide the mark-up necessary for automatically investigating
and understand the meaning of the data provided. For example, collections of
documents on the web are linked to each other without allowing inspecting the
motivation why these resources have been linked. The semantic web offers a variety
of new opportunities to overcome these shortcomings in today’s hypertexts. Amongst
others, the semantic web aims at enriching the content of the world wide web pages
with meaningful structure to create an environment in which software agents can
roam from page to page to carry out more intelligent tasks for their users [2]. Thereby,
the semantic web integrates (earlier) research from knowledge representation,
however, in the same decentralized manner the web is known for.
XML and RDF technologies constitute a basis for knowledge representation for
expressing web-based semantics. An important point is that XML just describes the
structure of data and does not say anything about its meaning, while RDF is used to
formalize semantics. However, RDF is not enough to describe complex relationships
between objects such as cardinality constraints, unions, disjoint classes etc. RDFS
gives a limited capability to define only simple relationships. At this point ontologies
come up as a solution, they are used to define complex relationships and sets of
inference rules between data objects by providing more vocabulary. There are also
many efforts for an ontology-based semantic web, important ones to mention here are
the DAML and OWL ontology languages. Both of these languages are based on
XML, RDF, and RDFS. XML, RDF, and OWL can be seen as the layers where each
layer requires different skills and targets different needs [3].
XML, RDF, OWL, and all of these technologies aim at expressing semantic in a
machine understandable way, but not to be understood by humans. It would be a big
burden to duplicate efforts by creating a piece of data in the form of both RDF, for
machines, and simple XHTML, for humans. Additionally, this would also be
problematic in terms of data consistency and synchronization. To avoid this double
work burden, eRDF, RDFa, and microformats have been proposed, allowing for
embedded semantic mark-up within web pages and being understandable for both
machines and humans. RDFa is a W3C specification based on expressing structure via
attributes of languages of XHTML and HTML [4]. eRDF, i.e. embedded RDF,
facilitates the embedding of crucial parts of the RDF model, but does not attempt to
extend this to the full RDF support [5]. eRDF is inspired by some basic principles of
microformats. A microformat is a set of XHTML tags that is used to embed
information into web pages which are understandable both by machines and humans
while considering the human as first priority [6].
RDFa and eRDF are based on the framework provided by RDF whereas
microformats offer both syntax and a restricted vocabulary that does not rely on RDF
or any other framework. Therefore, microformats are domain specific. A microformat
only becomes meaningful, when its syntax and vocabulary are defined by a
community. This also implies that the extracting procedure is same for every eRDF
and RDFa statement expressed in XHTML pages whereas it is different for every
microformat. eRDF and RDFa are based on RDF. They therefore enable users to mix
and use different name spaces. Microformat, however, use a flat name space that is
already predefined and cannot be extended or mixed; new metadata elements cannot
be added. It is obvious that RDFa and eRDF provide more flexibility and
functionality then microformats , however there are hardly any real life applications
of RDFa or eRDF whereas there are many real life examples of microformats.
Microformats do not aim at becoming a panacea for expressing taxonomies,
ontologies, or other abstractions which are not extensible or open ended [7].
While Microformats can encode explicit information to aid machine readability,
they do not address implicit knowledge representation, ontological analysis, or logical
inference [8]. The simplicity of this approach may be the reason of its success. Its
constraints, however, are the reason why it is called ‘lower case’ semantic web. A
living community being the key, it is not astonishing that most microformats express
well-established and accepted meta-data such normalized by Dublin Core, iCal,
vCard, FOAF, or ATOM. For instance, using the hAtom microformat, a website can
immediately provide a feed by intelligently marking up items. There is no longer a
need to publish the same content in other formats like RSS since applications can
extract raw ATOM data from the (X)HTML of the website, again without converting
to and from RDF [9].
Microformats are well accepted by now. Within the known limitations, they
provide already simple and straightforward capabilities to manipulate meaningful
structures added to human-understandable pages.
3 A Learning Object Harvesting Model for Web Content
The proposed model for harvesting learning objects bases assumes that learning
objects (also called ‘learning resources’) are digital and non-digital artifacts (images,
texts, etc.) that can be used for educational purposes. Learning objects can be part of
the courseware ranging in size and complexity from a single graphic to an entire
course [10]. Enabling learning objects to be re-usable and sharable by different tools,
systems, and applications is the core challenge for e-learning interoperability.
Hereby, meta-data plays a key role: there are several bodies which have published
standards and specifications to describe metadata information. In the context of this
work, LOM (Learning Object Metadata) standardized by IEEE is of prominent
interest, though other standards are available and compatible with the approach
chosen. Most of these standards consist of many elements with varying degrees of
relevance (depending mainly on the application and the usage context). For instance,
LOM currently offers more than 71 elements. From this standardized set offered by
LOM, an application profile has been derived mixing in elements from DC and
additional custom elements. The application profile forms the basis for a microformat
to embed learning resources into web pages that can later-on be harvested by
crawlers. Embedding microformat structures into XHTML can be realized by hand
with simple XHTML mark-up or it can be automated with the help of XSLT
transformations over the XML bindings of LOM.
Fig. 1. Harvesting model for extracting learning object meta-data from web-based content.
After proposing a light-weight microformat for learning objects, a web service
(based on SOAP and WSDL) has been created which harvests learning objects (bound
to microformat) in XHTML pages and transforms them into RDF format via XSLT or
XSLT and GRDDL [11] combined. If the XHTML file references an XSL file
providing transformation information on how to translate microformat bindings of
learning objects into RDF then GRDDL is used. Otherwise a predefined XSL
transformation is applied to perform the necessary transformation into RDF.
Additionally, an SQI target is provided which uses above mentioned harvesting
mechanism and allows querying for learning objects meta-data harvested into the
central storage facility. SQI is an interoperability infrastructure that enables
heterogeneous systems to communicate for the purpose of learning object retrieval by
using a common query language (in our case SPARQL) [12]. Fig. 1 provides an
overview on the workflow of the proposed harvesting and retrieval model. As a proof
of concept, a search application was implemented which uses the SQI target to query
learning resource meta-data.
4 Implementation Work and Experiences
Simplicity and a minimalist approach have been considered as the guiding mantra of
the prototypical implementation described below, especially while selecting data
types and meta-data elements. An e-Learning application profile guideline, which was
submitted to 2006 CEN/ISSS WS/LT workshop [13], has been followed to guide the
construction of the application profile (roughly consists of identifying the following
steps: requirement analysis, selection of data elements, multiplicity requirements, data
elements from multiple name spaces, local data elements, obligation of data elements,
value spaces, relationship and dependencies, and data type profiling). The primary
purpose of the proposed application profile is to facilitate re-usability and
interoperability of learning objects both by individuals and automated software
applications such as agents. The minimalist approach defines the borders of this
application profile by the vocabulary of our microformat. Therefore, it is important to
address common needs instead of providing a comprehensive profile, since schemas
of microformats are not subject to change often. Elements of the application profile
have been derived from IEEE LOM 1482-12-1 2002 and Dublin Core, enriched by
two custom elements. The application profile’s reference model and XML bindings
are based on SCORM 2004 3rd Edition.
The selection follows and extends the proposals for application profiles done by
Friesen and Campbell [14, 15] and later work of Godby [16] (which again is based on
Friesen and Campbell’s spreadsheets). A total of 35 application profile has been
investigated and lead to the conclusion that there is a close match between Dublin
Core elements and the frequently used IEEE LOM elements. Therefore we followed
the idea to map the 15 Dublin Core elements into the IEEE LOM elements if
available, which means we included only LOM elements having a respective Dublin
Core match, and if there is no such match, the Dublin Core element is taken as is. A
total of 18 elements have been proposed, seven of them adopted from the Dublin Core
name space, nine of them adopted from the IEEE LOM name space, and two
additional ones that have been derived from IEEE LOM through a custom name
space, this is done just in order to simplify those two aggregate elements. Tab. 1
provides an overview on the selected elements. Further details such as data types,
multiplicity requirements etc. have been omitted due to space limitations.
Tab. 1. Elements of the proposed microformat (LO: LOM, DC: Dublin Core, CO: Custom)
Set
LOM
DC
CO
Selected Elements
Title, Language, Description, Keyword, Coverage, Type, End User, Context, Format
Date, Creator, Contributor, Publisher, Rights, Relation, Source
Identifier, Classification
The proposal of microformat was quite straight forward in this case, because the
base applications profile itself is already simple and flat as there is no aggregate
element used in application profile at all. Most of the elements has been directly
associated with a ‘class’, i.e. class design pattern, attribute. The ‘creator’,
‘contributor’ and ‘publisher’ elements are used together with ‘vcard’ identifier. This
is because the domains associated with these elements are based on ‘vcard’. The abbrdesign pattern has been used for the ‘date’ and ‘language’ elements. The only
elemental microformat, i.e. the ‘rel-tag’ microformat, is utilized for the ‘keyword’
element. Fig. 2 depicts the extraction of the proposed microformat.
Fig. 2. Extracted meta-.data of an exemplary learning object encoded with our microformat.
An SQI target has been built and enhanced with GRDDL and SPARQL by using
the RAP RDF API [17]. Several sample microformat-enriched web-pages have been
generated which are available to the SQI target. Then, a simple search client which is
based on AJAX, JavaScript, CSS, and PHP has been set up to employ this SQI target,
after its harvesting the sample pages via GRDDL, to query microformat-encoded
semantics via SPARQL. These web-pages has been generated for test purposes,
however any individual or authoring tool supporting our proposed microformat is
capable of producing such microformat embedded pages by means of individual
authoring or by means of automated transformation of current content as far as a valid
match found between their previous (or base) metadata descriptions and our
microformat (this is usually the case if DC or LOM has been employed as base).
For instance the metadata for our sample learning objects are created with respect
to the XML LOM bindings. Thereafter, a simple XSL transformation embeds this
semantics into the web-pages according to our microformat automatically, and our
SQI target harvests this semantic information using GRDDL and transforms it into
RDF. Then, the query results are sent to search client in the form of a LOM XML
binding, and the search client visualizes the extracted XML via XSL. In Fig. 3 the
search results for the query term ‘elearning’ are displayed.
Fig. 3. Results for term ‘elearning’ visualized with our search client employing the SQI target.
It is important to highlight that the search client serves as a proof of concept,
particularly for SPARQL, i.e. the SQI target only employs SPARQL’s basic facilities.
Once the learning objects are represented with RDF or even empowered
ontologically, query languages for RDF like SPARQL will allow the application of
complex relationship queries and operations over learning objects in the sense of
retrieval, analysis, content construction etc.
Comparing our microformat-based to an ontology-based approach, it has to be
mentioned that microformats have clear shortcomings against RDF. Particularly,
microformats lack strengths like ‘open-ended design’, ‘the ontological expression
power’, and ‘extensibility’. However, a switch from Microformats to RDFa is in
principle possible, though at the time of writing not promising due to the widespread
acceptance and due to the ease-of-use of microformats. The drawbacks of
microformats constitute the main drawbacks of the model and the sample application.
5 Conclusion and the Future Work
Within this paper, we have introduced a model and a prototypical sample application
for learning object harvesting with the help of a new microformat in order to address
mainly interoperability and reusability of learning content. Apart from the drawbacks
which originate from the simplicity of microformats, this model needs to be
complemented by strong reasoning and data mining algorithms, and, additionally, by
techniques not just for harvesting objects but also for analyzing learning patterns and
learning activities described on the web. Efficient agents for crawling need to be
added as well. Those agents are not expected to serve just traditional desktop
computers, but should rather serve any device which has access to internet such as
PDAs or cell phones. In this way, learning could be embedded into real life.
References
1. Kieslinger, B., Wild, F., Arsun, O.I.: iCamp – The Educational Web for Higher Education.
In: Proc. of the EC-TEL, pp. 640--645 (2006)
2. Berners-Lee, T., Hendler, J., Lassila O.: The Semantic Web. Scientific American (2001)
3. Wilson, M., Matthews, B.: The Semantic Web: Prospects and Challenges. In: Proc. Of the
International Conference on Databases and Information Systems, pp. 26--29 (2006)
4. Adida, B., Birbeck, M.:
RDFa Primer. W3C (2008). Available
at
http://www.w3.org/TR/xhtml-rdfa-primer/#id85078 (Aug. 2008)
5. Talis:
Rdf
in
HTML:
Embedded
RDF
(2006).
Available
at
http://research.talis.com/2005/erdf/wiki/Main/RdfInHtml (Aug. 2008)
6. Allsopp, J.: Microformats: Empowering Your Markup for Web 2.0. FriendsofED, Berkeley
(2007)
7. Simpson, J.: Microformats vs. RDF: How Microformats relate to the Semantic Web (2007).
Available
at
http://www.semanticfocus.com/blog/entry/title/microformats-vs-rdf-howmicroformats-relate-to-the-semantic-web (Aug 2008)
8. Khare, R., Çelik, T.: Microformats: A pragmatic path to the Semantic Web. In: Proc. of the
International World Wide Web Conference, pp. 865--866 (2006)
9. Graf, A.: RDFa vs. Microformats. DERI Technical Report (2007)
10.Collier, G.: e-Learning Application Infrastructure. Whitepaper, Sun Microsystems (2002)
11.Connolly D.: Gleaning Resource Descriptions from Dialects of Languages. W3C (2004).
Available at http://www.w3.org/2004/01/rdxh/spec (Aug. 2008)
12. Prud'hommeaux E., Seaborne E.: SPARQL Query Language for RDF (2008). Available at
http://www.w3.org/TR/rdf-sparql-query/ (Jan. 2008)
13.CWA: Guidelines and support for building application profiles in e-learning (CEN
Workshop Agreement 15555:2006E), European Committee for Standardization (2006)
14.Friesen, N.: Survey of LOM Implementations, CanCore (2003)
15.Campbell, M.L.: UK LOM Core Update. CETIS Technical Report (2003)
16.Godby, C. J.: What do Application Profiles Reveal about the Learning Object Metadata
Standard? ARIADNE, October 2004 (41)
17.Melnik, S.: RAP: RDF API for PHP. SourceForge (2008). Available at
http://sourceforge.net/projects/rdfapi-php (Aug. 2008)