We present an experimental implementation of an ontological engineering support method that allows to see multiple OWL modeling options for a certain modeled situation. Initial style-independent background model (PURO OBM) is automatically transformed to alternative OWL fragments, which can then be visualized and chosen from. During ontology engineering in OWL, the developer can often choose di erent combinations of language constructs to model the same situation. The choice might be driven by the intended usage of the ontology: web markup vocabularies often favor `feature' assignment to entities through data properties, while linked data vocabularies prefer object properties for this purpose; reasoning-enabled ontologies, in turn, express `features' as classes. We call consistent sets of such choices modeling styles. Viewing di erent alternatives may help supporting the choice of style for a speci c use case. We propose a method where a single styleindependent model { ontological background model (OBM) in PURO language [6] { is created manually, and the style-speci c OWL representations are then automatically generated using OBM2OWL transformation patterns. While not being a way to author an ontology in bulk, this approach can serve as `spotlight' on di erent modeling options. PURO OBMs are based on two main distinctions: between particulars and universals and between relationships and objects (hence the PURO acronym). There are six basic entity types: B-object (particular object), B-type (type of object/type), B-relationship (particular relationship), B-relation (type of relationship), B-valuation (particular assertion of quantitative value) and B-attribute (type of valuation). A PURO OBM consists of named instances of these types, plus of subTypeOf and instanceOf relationships. It captures a concrete situation that serves as basis for analyzing existing ontologies or generating new ones.
Related research Compared to heavier-weighted OBM formalisms such as
OntoUML [
Another example: to say that `025.04' is a topic in the Dewey Decimal Classi cation (DDC), at least two ways are possible in OWL: { class instantiation: o:DDC Topic rdfs:subClassOf d:Topic.
d:025.04 a o:DDC Topic. (cf. British Library Data Model)
{ dedicated `type' class + object property: o:ddc topic a o:TopicType.
d:025.04 o:topicType o:ddc topic. (cf. the openlibrary.org dataset)
Rather than aligning such choices pairwise, as we did in our previous project,
PatOMat [
The OBM2OWL transformation patterns are currently implemented as SPARQL
UPDATE queries, with structure based on OWL2OWL patterns from PatOMat.
The SELECT part describes a fragment of OBM, e.g., a B-relationship
connecting two B-objects. The INSERT part describes a corresponding OWL structure.
We so far created six patterns4 covering rather intuitive transformations. The
transformation functionality was embedded into PURO Modeler,5 our web-based
graphical editor of OBM. Once the user creates an OBM, it can be serialized
into RDF and sent to a web service that executes the SPARQL queries and
returns OWL ontology fragments separated into RDF graphs. Each graph is then
1 http://www.w3.org/2001/sw/BestPractices/OEP/
2 Based on a survey from http://tomhanzal.github.io/owl-modeling-styles/.
3 The \d" pre x always refers to a ctional dataset and \o" to a ctional ontology.
4 Available at http://lod2-dev.vse.cz/puromodeler/patterns/.
5 http://lod2-dev.vse.cz/puromodeler
visualized using WebVOWL [
Consider, again, a designer of a bibliography ontology who wants to model book topics and place of publishing. In PURO Modeler s/he creates/opens the OBM of an example situation with book1 about DDC topic 025.04 published in Prague (Fig. 1). The tool serializes the OBM into an RDF graph6 and returns four alternative OWL ontology fragments from Fig. 2, which combine the style options described in Section 1. In fragment A, object properties are used where possible: even the topic type assignment is modeled as a property (topic hasType). In B, both relationships are rei ed (hasTopic Relationship, publishedIn Relationship) and the topic type assignment becomes class instantiation. C is similar to A, but topic type assignment is modeled as in B. In D, all relationships are modeled as data properties; the topic type is omitted since the topic is now a literal. The designer can now choose the most suitable OWL fragment, download it and integrate into the ontology using his/her favorite editor.
In the future, generation of OWL fragments could be restricted to those complying with best practices; e.g., lexical encapsulation (BookPublishednInPrague) could be omitted as suboptimal, requiring a new class for every location. For some OBM structures the number of variants will also be smaller by de nition, e.g., an n-ary relation with n>2 cannot be modeled as plain OWL property. The presented approach of ontology style bootstrapping from an OBM is still in its infancy and the initial set of OBM2OWL patterns is merely tentative; yet, thanks to tool availability, feedback can already be collected. The longterm vision is that of a portal (integrated with ontologydesignpatterns.org ) where 6 Available at http://lod2-dev.vse.cz/puromodeler/bookLocationTopicOBM.rdf experts trained in OBM (`PURO writers') could build and share the OBMs while novice ontological engineers (`PURO readers') could pick fragments of the OBMs and let the OBM2OWL service generate the corresponding OWL fragments for them. Systematic elaboration and analysis of advantages and disadvantages of the modeling styles is planned for future research. Use of lexicographic resources to generate sound labels for new elements is also foreseen.
The research is supported by UEP IGA F4/90/2015 and by long-term institutional support of research activities by Faculty of Informatics and Statistics, Univ. of Economics, Prague. Ondrej Zamazal is supported by CSF 14-14076P.