There are a plethora of techniques for carrying out ontology module extraction, see Section 3.4. All of these techniques are designed for different applications and contain different assumptions about the problem as whole. Thus, there is a need to draw this work together within a common framework; the key advantage of a common framework is the ability to select, adapt and combine the different approaches. This would greatly facilitate an objective evaluation and comparison of the different ontology modulariza- tion techniques; Chapter 4 presents an in-depth evaluation of ontology modularization, including a metric and task based evaluation.
Also, the development of new techniques is made easier, since common technical issues, such as whether reasoning is used or not, are already tackled in the framework and little effort is required by the developer.
3.6.1 Tell/Ask Interface
Borgida and Giunchiglia [8] present a Tell/Ask interface for ‘importing knowledge’, that is reusing knowledge from existing ontologies. This work can be viewed as a common framework for ontology module extraction. Inspired by Levesque’s functional approach to knowledge representation [87], which allows the user to interact with a knowledge base via Tell and Ask operators. You ‘Tell’ the TBox facts and then ‘Ask’ the TBox queries about the facts it contains. Borgida and Giunchiglia cast this to Description Logics. In this context, Tell operations allow a TBox to be built and Ask operations allow the knowledge base to be interrogated.
Therefore, this work can be cast to an approach for a common framework for on- tology module extraction. A series of Ask operations would simulate the extraction approach. The answers of these operations would form Tell operations to construct a new TBox, the ontology module. This allows for a flexible framework, but adding new operations is likely to be costly as each will have to be implemented separately. 3.6.2 Graph Transformations
d’Aquin et al. [26] suggest graph transformations as a possible common framework. Firstly, d’Aquin et al. [26] present a way to transform ontologies into a directed at- tributed graphs, which is a directed graph where attributes, in terms of types and values can be added to the nodes and edges. Thus, a node representing a concept C
can have typeClass and valuename= C. For example, the axiom
P ersonW ithDogAndCat≡P ersonu ∃hasP et.Dogu ∃hasP et.Cat
would be transformed into the graph shown in Figure 3.2
The existing techniques ( [27], [33], [103] and [115]) are then represented as a series of graph transformations. A graph transformation takes one graph as a pre-condition
N1:Class (name=P ersonW ithDogAndCat)
N2:Class(const=u) N3:Class (name=P erson)
N4:Class (const=u) N5:Class(const=∃)
N6:Class (const=∃) N7:Class (name=Dog)
N9:Class(name=Cat) N8:P roperty (name=hasP et)
≡
op2
op2
someV aluesF rom op1
op1
p someV aluesF rom
p
Figure 3.2: Attributed graph representation of the expression
P ersonW ithDogAndCat≡P ersonu ∃hasP et.Dogu ∃hasP et.Cat
and one graph as a post-condition. For example, Figure 3.3 shows the graph transfor- mation for the downwards traversal of the subclass hierarchy. This allows for an easily extensible framework, but graph transformations are not widely applied in the Ontol- ogy Engineering and Semantic Web communities, thus there could be a considerable user overhead in learning how they operate.
Premiss Transformation C1:Class(inc) C2:Class v C1:Class(inc) C2:Class(inc) v(inc)
Figure 3.3: An example graph transformation for downwards traversal of a subclass hierarchy.
3.6.3 SPARQL Based Extraction
The work by Borgida and Giunchiglia [8] and d’Aquin et al. [26] require the user to become familiar with non-standard formalisms, but the work by Doranet al. [32] uses the W3C standards of RDF and SPARQL as the basis for a common framework for ontology module extraction. All OWL ontologies can be represented as an RDF graph (see Section 2.4) and SPARQL is a query language for RDF. Thus, Doran et al. [32] present SOMET which shows that it is possible to cast the traversal based ontology module extraction approaches as a series of SPARQL queries upon an RDF graph. Thus, the selection, adaptation and combinations of the techniques are manipulations of SPARQL queries.
The SOMET framework (see Figure 3.4) already includes the SPARQL representa- tions of the techniques presented in [27], [33], [103] and [115]. For example, some of the queries required for Doranet al.’s [33] technique, where ?c is the current concept
Traversal Extraction Engine Ontology Module Ontology SPARQL Queries
PROMPT Galen D'Aquin Doran
Signature
Figure 3.4: The SOMET framework. [32] of focus, are:
• DESCRIBE?c
Describes the current resource. That is the list of the statements in which the ?c.
appears as subject, plus the closure computed from any blank nodes involved.
• CONSTRUCT{?y rdfs:domain?c.}WHERE{?yrdfs:domain?c.}
Returns all the ?y where ?c.is the domain of a property.
• DESCRIBE?yWHERE{?yrdfs:subClassOf?c.}
Returns all the subclasses of ?c.
• CONSTRUCT{?y owl:equivalentClass?c.}WHERE{?yowl:equivalentClass?c.}
Returns all the ?y where ?y is an equivalent class to ?c.
Furthermore, the framework allows these queries to be modified or new ones to be added. Thus, the framework is fairly flexible and allows for experimentation in a formalism that is standard and utilised within the Semantic Web community.