=Paper=
{{Paper
|id=Vol-1299/paper3
|storemode=property
|title=What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler
|pdfUrl=https://ceur-ws.org/Vol-1299/paper3.pdf
|volume=Vol-1299
|dblpUrl=https://dblp.org/rec/conf/ekaw/DudasHS14
}}
==What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler==
https://ceur-ws.org/Vol-1299/paper3.pdf
What Can the Ontology Describe?
Visualizing Local Coverage in PURO Modeler
Marek Dudáš, Tomáš Hanzal, and Vojtěch Svátek
University of Economics, Prague,
marek.dudas|xhant00|svatek@vse.cz
Abstract. Ontologies and vocabularies written in OWL are a crucial
part of the semantic web. OWL however allows to model the same part
of reality using different combinations of constructs, constituting ‘mod-
eling styles’. Comparing how different ontologies from a similar domain
cover a specific part of reality might be more difficult when each ontology
uses a different style. PURO, a language for ontological background mod-
els, can serve as mediator, as it allows to create models that are directly
mappable to OWL but overcome its unnatural modeling limits dictated
by description logic. By highlighting the parts of the PURO model cov-
ered by particular ontologies, the ontology coverage comparison can be
visualized. We demonstrate this approach using a simple graphical tool.
1 Introduction
OWL [8], the language of ontologies on the semantic web, allows to model the
same real-world state of affairs using different combinations of language con-
structs, a kind of modeling styles. For example, in a lightweight ontology, data
properties might be preferred, while in a more complex ontology, object proper-
ties might be used for describing the same relationships, allowing to state more
facts about the property value. Current ontology visualization tools (as surveyed
in [4,6]) either display the OWL constructs directly or only offer very lightweight
generalization, as in [5]. When the user wants to compare, in a visual manner,
the local coverage 1 of different ontologies, i.e. their capability to express a certain
cluster of relationships, s/he has to first translate the OWL language constructs
into his/her mental model to abstract from the modeling differences.
A possible solution in such a situation is to express the above mentioned
cluster of relationships in a modeling language that allows to abstract from the
modeling style differences and thus make the mental model explicit. We believe
that PURO ontological background models (OBMs) [9] might fit this use case.
We are developing PURO Modeler, a Javascript-based tool for graphical design
of OBMs and ontology local coverage comparison.2 In the paper, we demonstrate
its usage on two examples.
1
We assume that typically only a few entities and relationships are considered from
the analyzed models (e.g., a book, its author and its subject). To differ from the
coverage of the whole domain (e.g., bibliography), we use the term ‘local coverage’.
2
The development version is available at http://protegeserver.cz/puromodeler.
28
What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler
Related Research We are unaware of any prior research on visual comparison of
ontology local coverage. Creating and visualizing mappings [3] between the com-
pared ontologies might be useful but it only shows what the ontologies have in
common. There is research on ontology comparison (e.g., [7]), but that is rather
focused on automatic computation of ontology similarity. In our approach the
actual analysis is left to the user; we however propose a method and means for
supportive visualization. The PURO language, designed to be easily mappable
to OWL, is used as interlingua for this purpose. Other modeling languages, e.g.,
entity-relationship diagrams [2], might be considered as alternatives to PURO.
Of those, OntoUML [1], a version of UML for conceptual modeling, is probably
the closest match; an existing tool, OLED, even allows to transform OntoUML
models into OWL fragments. However, the purpose of OntoUML is conceptual
modeling with easy validation and it is not designed for OWL ontology develop-
ment or usage.
2 PURO Ontological Background Models
The long-term vision of the PURO OBM language is to model the real world,
as much as possible, ‘as it is’, while remaining ‘very close’ to OWL. It is only
intended to serve as an aid in ontological engineering, i.e. models created in it
are not supposed to serve (in large scale) as schema for data or input to reason-
ers. The liberation from the constraints dictated by description logic allowed to
drop some constraints of OWL (DL, or even Full) that are often perceived as
unnatural; most notably, meta-concepts (such as classes of classes, or classes as
‘property values’) and arbitrary n-ary relationships can be expressed in PURO.
PURO OBMs are based on two main distinctions: between particulars and
universals and between relationships and objects (hence the PURO acronym:
Particular-Universal, Relationship-Object). In simple terms: universals, i.e. types,
can have instances, while particulars cannot (P-U distinction); objects and rela-
tionships are differed by their identification: objects are entities with their own
identity that can be ‘talked about’ independently, while talking about relation-
ships always ‘brings in’ their participating entities (R-O distinction). There are
six basic terms: B-object (particular object), B-type (type of objects/types),
B-relationship (particular relationship), B-relation (type of relationship), B-
valuation (particular assertion of quantitative value) and B-attribute (type of
valuation).
An OBM consists of named instances of these terms and relationships be-
tween them.3 To say, it captures a concrete situation that serves as basis for
either analyzing existing ontologies or generating new ones.
3 Using PURO Modeler for Local Coverage Comparison
PURO Modeler is basically a graph editor. The user can create an OBM where
the terms are represented by nodes and the relationships between them by edges.
3
The description of the language is simplified, more details are in [10].
29
What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler
As the OBMs are models of specific real-world situations, such an example cluster
of relationships has to be chosen. After constructing the model,4 or loading an
already existing one (possibly created by another user and shared), the user
can select each node and annotate it with the ontologies that this particular
instance of PURO term is covered by (i.e., the instance can be described with
that ontology). The parts of the graph locally covered by each vocabulary are
then highlighted so that the coverage could be easily compared in one diagram.
3.1 Example 1: A Book and its Issue
Let us say we need to annotate data about books and their various issues (i.e.
their manifestations). We can find several ontologies that might be used for it.
BIBFRAME5 contains class Text as a subclass of Work, class Print as a sub-
class of Instance and the hasInstance property for stating that an instance of
Print is a manifestation of some Text. Figure 1b shows an example of possible
usage of BIBFRAME to describe the book The Semantic Web for the Work-
ing Ontologist and its printed paperback manifestation. FRBR ontology6 (and
its extension Fabio7 ) distinguishes between four different ‘levels of abstraction’
represented by classes Work, Expression, Manifestation and Item, each with sub-
classes representing more specific types. There is a set of properties that can be
used to link the instances of the classes, as shown in Figure 1c. Schema.org
does not distinguish between the work and its manifestation through class mem-
bership, but contains properties exampleOfWork and workExample that serves
that purpose. It contains the class Book and allows to specify the format of
its instances by linking them to instances of BookFormatType with property
bookFormat (Figure 1a).
For the purposes of the local coverage comparison, we chose the book The
Semantic Web for the Working Ontologist (as an instance of the B-type Text,
which is a subtype of Work) and its 2009 paperback issue (an instance of the B-
type Paperback, a subtype of Book). An OBM of these two entities accompanied
by a B-object representing an exemplar of the paperback issue (as a tangible
item) and the relationships between them is in Figure 2 (exported as a screenshot
from PURO Modeler). The color-coded highlighted parts of the model shows
what of it can be expressed in each vocabulary.
We can see that, e.g., the physical exemplar of the issue can only be de-
scribed with FRBR (which ’implements’ the whole OBM), BIBFRAME allows
to describe an issue of a book, but without specifying its format, and Schema.org
does not allow to say that an instance of work is a Text.
4
The methodology for the OBM modeling is yet to be formalized.
5
http://bibframe.org/vocab
6
http://purl.org/vocab/frbr/core#
7
http://purl.org/spar/fabio/
30
What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler
Fig. 1. Diagrams showing three different OWL modeling styles of the relationship
between a book and its issue.
Fig. 2. The OBM of the relationship between a book and its issue. The amount of its
coverage in various OWL representations (shown in Figure 1) is highlighted.
31
What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler
3.2 Example 2: A Dish and a Recipe
In the second example, we compare the local coverage of the relationships be-
tween ingredients, a recipe and a dish produced by it. We analyzed8 three ontolo-
gies: Schema.org, Food Ontology9 and Linked Recipes.10 Their local coverage is
highlighted in Figure 3 (for simplicity, we made up a ’boiled egg recipe’ with only
one ingredient and we did not include all possible B-attributes of the B-objects,
like ’protein content’ of the dish).
We can see that Schema.org allows to describe the recipe in terms of its
origin, ingredients and time needed; the carbohydrates, fat and energy content;
and size of the produced food. It does not allow to model the recipe instructions:
there is only one datatype property for entering a free text description. Linked
Recipes does not allow to describe the characteristics of the recipe product,
but (unlike Schema.org) allows to type it as ’Food’. It can be used to annotate
instructions including their order, and ingredients including their quantity. Food
Ontology can only be used for the (quite detailed) description of the ingredients
and the resulting dish. It does not contain classes/properties relevant for cooking
instructions. None of the three ontologies covers the whole situation as modeled
by the OBM, however, we can see that a combination of two of them might
be sufficient, e.g., using Linked Recipes for recipes in combination with Food
Ontology for more detailed ingredient and dish description.
4 Conclusions and Future Work
We argued that the task of ontology comparison in terms of local coverage is
difficult if the ontologies use different modeling styles. We proposed that using
a more general modeling language for visualizing the local coverage of several
ontologies in one place, abstracting from the OWL modeling differences, might
make the comparison easier, and that the PURO OBM language might be suit-
able for such a use case. As a first step towards the evaluation of the approach we
implemented PURO Modeler: a Javascript application for PURO OBM models
design and visualization of the amount of their local coverage in various ontolo-
gies. We demonstrated the usage of PURO Modeler for local coverage comparison
on two examples, each consisting of an analysis of three ontologies.
The future work might include, besides the full evaluation, building a portal
where the developers could publish visualizations of their ontology/vocabulary
local coverage of typical situations. Such a website could then serve as an addition
or enhancement of the existing Linked Open Vocabularies11 portal. Intense on-
going research also addresses the obvious bottleneck of the approach: the design
of plausible PURO OBMs. It is going to be supported by interactive guidelines
and NLP-based model verbalization widgets.
This research is supported by the VŠE IGA project no. F4/34/2014.
8
See http://tomhanzal.github.io/owl-modeling-styles/#part2 for full analysis.
9
http://data.lirmm.fr/ontologies/food/
10
http://linkedrecipes.org/schema/
11
http://lov.okfn.org/dataset/lov/
32
What Can the Ontology Describe? Visualizing Local Coverage in PURO Modeler
Fig. 3. The OBM of a recipe, ingredients and dish. The amount of its coverage in
various OWL representations is highlighted.
References
1. Albuquerque, A., Guizzardi, G.: An ontological foundation for conceptual modeling
datatypes based on semantic reference spaces. In Research Challenges in Information
Science (RCIS), 2013 IEEE Seventh International Conference, 2013.
2. Chen, P.: The entity-relationship model – toward a unified view of data. In: ACM
Transactions on Database Systems (TODS), 1976, 1.1: 9-36.
3. Choi, N., Song, I. Y., Han, H.: A survey on ontology mapping. ACM Sigmod Record,
35(3), 34-41. 2006.
4. Dudáš, M., Zamazal, O., Svátek, V.: Roadmapping and Navigating in the Ontology
Visualization Landscape. Accepted to EKAW 2014.
5. Hayes, P., et al.: Collaborative knowledge capture in ontologies. In: Proceedings of
the 3rd international conference on Knowledge capture – K-CAP 2005, ACM, 2005.
6. Katifori, A., et al.: Ontology Visualization Methods – A Survey. ACM Comput.
Surv. 39, 10143. 2007.
7. Maedche, A., Staab, S.: Measuring similarity between ontologies. In: Knowledge
engineering and knowledge management: Ontologies and the semantic web. Springer
Berlin Heidelberg, 2002. p. 251-263.
8. OWL 2 Web Ontology Language Document Overview. W3C Recommendation, 2009.
9. Svátek, V., et al.: Mapping Structural Design Patterns in OWL to Ontological
Background Models. In: K-CAP 2013, ACM, 2013.
10. Svátek, V., et al.: Metamodeling-Based Coherence Checking of OWL Vocabulary
Background Models. In: OWLED 2013, online http://ceur-ws.org/Vol-1080/
owled2013_6.pdf.
33