The use of ontologies is a key step forward for describing the semantics of information on the Web. It is becoming more and more important to make the information machinereadable, since the volume of data is continuously growing. In the educational area, metadata are considered to be helpful in such a process. We propose to enrich the description of educational resources by introducing several levels of description of concepts, and to make them machine- readable by using a formal language of ontology, OWL. Using both this ontology and the expressive power of an OWL query language to query pedagogical resources will improve the retrieval and interchange of educational resources.
Educational resources available on the Web are intended
to be shared, accessed or reused. Because of the ambiguity of
the natural language (synonymy, polysemy, homonymy,
multilingualism) the answers are spoilt by noise. Actually,
keywords of the query are matched with indices extracted
from the Web pages, but neither the semantics nor the
structure are taken into account by the search tools. Some
solutions have been proposed in order to explain the
semantics of the Web: we note the recommendation of
metadata Dublin Core [
The pedagogical ontology concerns both the organization of an engineer school: Supélec and the content of a teaching program.
We first present a part of a UML model of the teaching organization at Supélec.
Education
Teaching_Program
At Supélec, the education lasts three years. Each year is
divided into four sequences and contains several teaching
modules (each module corresponding to one course per
sequence). A module contains learning resources which are
either atomic or composite. In the LOM terminology, a
learning object is considered as a learning resource, this
equivalence can be expressed with an OWL restriction. The
UML schema can be transformed into an RDF representation
by the way of XPetal [
The domain and scope of the second part of the pedagogical ontology are the learning resources participating in a teaching program, created by teachers or educational organizations. In order to preserve the semantics given by the LOM, we mention some definitions: Learning Object: any entity that may be used for learning, education or training.
Category: a group of related data elements.
Data element: a data element for which the name, explanation, size, ordering, value space and datatype are defined in the LOM standard.
Now we present two other UML representations of a learning objects view and a LOM metadata view.
Text Video
Module title duration Learning_Resource name description Image
Sound
RawData
contains 0..* Learning_Object
0..* 0..* owns isDescribed 1..* Complex
Atomic Bag
Set
List
Graph
In figure 2, the Module and Learning resources are two types of Learning Objects in the terminology of LOM. A learning object is composed of raw data, media, structure and metadata. The media is text, sound, image or video. Each media type has a format (for example jpeg for an image, MP3 for a sound). The structure of a learning object is either atomic or complex (for example a definition, an example or a theorem is an atomic learning object whereas a module of software engineering is a complex one). Each learning object is described by a set of metadata which are detailed in figure 3. This representation reflects the view of LOM metadata with the concepts of categories, data elements and types of data element (the structure to represent the logical relationships between learning objects and the content to represent the content of a learning resource).This is an extract of the OWL representation of the learning object’s view.
<owl:Class rdf:ID="Text"> <owl:disjointWith> <owl:Class rdf:about="#Video"/> </owl:disjointWith> <rdfs:subClassOf> <owl:Class rdf:about="#Media"/> </rdfs:subClassOf> </owl:Class> <owl:Class rdf:ID=”Bag”> <rdfs:subClassOf rdf:resource=”#Complex”/> </owl:Class> <owl:Class rdf:ID=”Set”> <rdfs:subClassOf rdf:resource=”#Complex”/> </owl:Class> <owl:Class rdf:ID=”List”> <rdfs:subClassOf rdf:resource=”#Complex”/> </owl:Class> <owl:Class rdf:ID=”Graph”> <rdfs:subClassOf rdf:resource=”#Complex”/> </owl:Class> </owl:Class> <owl:class rdf:ID=”Bag”> <owl:disjointWith rdf:resource=”Set”/> <owl:disjointWith rdf:resource=”List”/> <owl:disjointWith rdf:resource=”Graph”/> </owl:Class>
DataElement RelationShip
Content
Value Structure
Semantics hasPart isReferencedBy isSimilarTo
A domain view (for example a thesaurus of the computer science) is illustrated by a hierarchy of terms, that guaranty there is no ambiguity in terms of understanding. The following extract of the classification of computer science built by ACM can be also translated into OWL: D SOFTWARE D.0 GENERAL D.1 PROGRAMMING TECHNIQUES (E) D.2 SOFTWARE ENGINEERING (K.6.3) D.2.0 General (K.5.1) D.2.1 Requirements/Specifications (D.3.1)
Elicitation methods (e.g., rapid prototyping, interviews, JAD) (NEW)
Languages Methodologies (e.g., object-oriented, structured) (REVISED) Tools D.2.2 Design Tools and Techniques (REVISED) D.2.3 Coding Tools and Techniques (REVISED)
Object-oriented programming (NEW) D.2.4 Software/Program Verification (F.3.1) (REVISED) Assertion checkers
Class invariants (NEW) D.2.5 Testing and Debugging
Testing tools (e.g., data generators, coverage testing) (REVISED)
Tracing This hierarchy of terms may be represented in OWL with subclass and equivalent relations. <owl:Class rdf:ID=”D”>
<rdfs:label>Software </rdfs:label></owl:Class> <owl:Class rdf:ID=”D2”> <rdfs:label>Software engineering </rdfs:label></owl:Class> <rdfs:subClassOf rdf:resource=”#D”/> </owl:Class> <owl:equivalentClass rdf:resource=”#K.6.3”/> </owl:Class> </owl:Class>
Each of the view was translated and refined by the
OWL formalism.
2.2. Description of learning resources with
the LOM semantics
The preliminary task consisted in translating the model of the
LOM into a schema in OWL. We did it with the Protégé
2000 editor [
D, R
D, R dc :hasPart
Relation S dc :isPrerequ isite
lom:Educational.
LearningResourceType.
lom :Graph General.AggregationLevel = 1, or that a teaching program has value General.AggregationLevel > 2.
Let’s go further with the composition of learning objects which has been evoked in sections 2.1 and 2.2.
We define two categories of links among learning objects:
the structural and the semantical ones. The structural links
correspond to the logical structure of resources (“hasPart”
and “sequence” links) whereas the semantical links
correspond to the semantics of the associations among
resources (besides the various relations defined in Dublin
Core [
Figure 5. simplifies an example of the description of learning resources with two levels of representation: the schema level and the instance one. The schema level is described thanks to an ontology, the instance level is the knowledge base. dc:haspart
R
T &r2 owl :oneOf dc:subject O :UML rdf :Bag T owl :oneOf O :RUP &r3 &r4 dc:subject
Program T dc:subject &r1
OWL-QL is a formal language and it is intended to be a candidate standard language for query-answering among semantic web computational agents. An OWL query contains a query pattern that specifies a collection of OWL sentences in which some URIrefs are considered to be variables. These answers provide bindings of URIrefs or literals to some of the variables in the query. For example, we could ask “Is there any course module whose the author is Mike?” The query can have the form: “(type ?c module) (author ?c mike)” where each query pattern is represented by a set of triples of the form (property subject object) and the variables are prefixed by the character “?”. Inference mechanisms enable to deduce new information from some properties (symmetry, transitivity…).The OWL language allows us to specify property characteristics, which provide a powerful mechanism for reasoning about a property. The property can be exploited in the query part. For example consider the transitive property in OWL. If a property P is specified as transitive then for any x, y and z: P(x,y) and P(y,z) implies P(x,z). The structural relation “isPartOf” is transitive. This allows us to define simple query for asking about any learning object linked to a module by a direct or indirect “isPartOf “ structural relation : Q1 : (type ?x LearningObject) ( type ?y module) (isPartOf ?x ?y) From the example Figure 5., finding a graph resource illustrating a module or a course in software engineering is expressed with an OWL-QL like language as: Q2: (type ?c Graph)(or (isPartOf ?c module) (isPartOf ?c course)) Finding all semantical links related to the &r1 resource: Q3 : (type ?c SemanticalRelation) (rdf:Range ?c &r1) As we illustrated in some of these examples, possibilities in expressing various powerful queries widen from schema and data queries, metadata, structural and semantic links, and reasoning forms.
In the building of a pedagogical ontology at Supélec, we distinguished two domains. The first one represented an education’s organization. It has been enriched with the second ontology which represented a pedagogical content using standardized metadata (LOM). We showed the importance of the relations among learning objects to infer additional knowledge in the querying step. We use Protégé 2000 for our examples. We edited the entire LOM schema, the schema and instances of our pedagogical ontology. Protégé 2000 enabled us to detect and solve some inconsistencies in the classes and relations and therefore to validate our schema. It is possible to query some simple facts and to make some inferences. We gave some examples of queries in order to show the expressive power of a query language exploiting the benefit of the ontologies. OWL-QL syntax was chosen to show some examples of queries but we consider other query formalisms. We are currently implementing an OWL query language to test further our pedagogical ontology.