A variety of ontology visualizations have been presented in the last couple of years. The features of these visualizations often need to be tested during their development or for evaluation purposes. However, in particular for the testing of special concepts and combinations thereof, it can be difficult to find suitable ontologies. We have developed OntoViBe, an ontology covering a wide variety of OWL 2 language constructs for the purpose of testing ontology visualizations. We describe the design principles underlying OntoViBe and present the supported features in coverage matrices. Finally, we load OntoViBe with ontology visualization tools and point to some noteworthy aspects of the respective visualizations that become apparent and demonstrate how OntoViBe can be used for testing ontology visualizations.
Developing and working with ontologies can be supported by ontology
visualizations. Over the past years, a number of visualization approaches geared
towards the peculiarities of ontologies have been proposed. Most of the available
approaches use node-link diagrams to depict the graph structure of ontologies,
while some apply other diagram types like treemaps or nested circles [
During the development of such ontology visualizations, testing with a variety of existing ontologies is required to ensure that the concepts from the underlying ontology language are adequately represented. The same needs to be done to determine the features of an ontology visualization and get an impression of how different ontology language constructs are visually represented. Still, repeatedly loading a set of ontologies that cover a wide variety of language constructs can be a tedious task, even more so as the most common constructs tend to appear over and over in each of the tested ontologies. In order to help that process with respect to ontologies based on the OWL 2 Web Ontology Language, we developed OntoViBe, an Ontology Visualization Benchmark.
Basically, OntoViBe is an ontology that has been designed to incorporate
a comprehensive set of OWL 2 language constructs and systematic
combinations thereof. While it is oriented towards OWL 2, it also includes the concepts
of OWL 1 due to the complete backwards compatibility of the two ontology
languages, i.e., all OWL 1 ontologies are valid OWL 2 ontologies [
As opposed to most other benchmarks found in the computing world, OntoViBe is not meant for testing the scalability of visualizations with respect to the number of elements contained in ontologies, but rather aims for the scope of visualizations in terms of supported features. Related to this, it focuses on the representation of what is usually called the TBox of ontologies (i.e., the classes, properties, and datatypes), while it does not support the testing of ABox information (i.e., individuals and data values), which is the focus of most related work. 2
Several benchmarks for ontology tools have been developed in the past. One
wellknown benchmark in this area is the Lehigh University Benchmark (LUBM),
published by the SWAT research group of Lehigh University [
Since the LUBM benchmark is bound to the university domain, the SWAT
research group developed another benchmark that can be tailored to different
domains [
All these benchmarks focus primarily on performance, efficiency, and
scalability, but do not address the visual representation of ontologies. Furthermore,
they are mainly oriented towards instance data (the ABox), while systematic
combinations of classes, properties, and datatypes (the TBox) are not further
considered. Even though UOBM provides comparatively complete TBox
information, it has been designed to test OWL reasoners and not ontology
visualizations. This is also the case for JustBench [
There are also some benchmarks addressing specific aspects of ontology
engineering. A number of datasets and test cases emerged, for instance, as part
of the Ontology Alignment Evaluation Initiative (OAEI) [
The W3C Web Ontology Working Group has also developed test cases for
OWL 1 [
The structure and content of OntoViBe is based on the OWL 2 specifications [
OntoViBe was assembled by creating an instance of each of the subgraph structures. Where possible, classes were reused to keep the ontology small. For instance, to include the OWL element object property, a subgraph structure consisting of two classes connected by an object property was added. Hence, two classes were inserted into the ontology, and an object property that uses either of the two classes as its domain and range, respectively, was defined. Furthermore, the element datatype property needed to appear in the ontology. A compact subgraph structure to express that element consists of a class linked to a datatype property. As the class does not need to have any specific characteristics of its own, one of the two previously inserted classes could be reused.
After that, some of the existing elements were modified to cover all elements and features that we wanted to consider at least once in the ontology. For example, some properties were declared as functional or deprecated. For any element type that still did not appear in the ontology, a minimal number of extra classes were added (the addition of extra properties and datatypes was not necessary). Listing 1.1. Concepts based upon set operators are featured in two variants, a small set with two elements, and a larger one with more elements. 30 this : UnionClass a owl : Class ; 31 owl : unionOf ( this : Class1 this : DeprecatedClass ). 32 33 this : LargeUnionClass a owl : Class ; 34 owl : unionOf ( this : UnionClass other : ImportedClass this : PropertyOwner ).
Listing 1.2. OntoViBe defines custom OWL data ranges. 94 this : DivisibleByFiveEnumeration a rdfs : Datatype ; 95 owl : equivalentClass [ 96 a rdfs : Datatype ; 97 owl : oneOf ( 5 10 15 20 ) 98 ]. 99 100 this : UnionDatatype a rdfs : Datatype ; 101 owl : unionOf ( this : DivisibleByTwoEnumeration this :
DivisibleByFiveEnumeration ).
Lastly, all elements in the ontology were named in a self-descriptive manner to allow for an easier interpretation and analysis. For instance, a deprecated class is called DeprecatedClass, while the larger of the union classes is called LargeUnionClass. 3.1
Many of the structures could be added in a straightforward way. In some cases, further considerations were required to adequately address the more flexible features of OWL.
Concepts defined based upon set operators (unionOf, intersectionOf, complementOf ) come in two variants. One of them uses a set comprising two elements as an example for a small set, while the other features more set elements, usually three (Listing 1.1).
OntoViBe also includes OWL data ranges (Listing 1.2). Visualizations may or may not represent the exact definitions of these data ranges, but even if they do not, support for datatype properties with custom data ranges needs to be tested. Therefore, custom data ranges are used by some datatype properties in OntoViBe, while common datatypes are used for most other properties (Listing 1.3).
In order to check how imported ontology elements are treated, OntoViBe consists of two components. The core ontology1 contains most of the definitions, but a few classes, properties, and datatypes are defined in an additional module2, whose content is imported into the core ontology (Listing 1.4). 1 http://ontovibe.visualdataweb.org/1.0# 2 http://ontovibe.visualdataweb.org/1.0/imported#
Listing 1.3. Both custom and common datatypes are used by properties. 114 this : standardTypeDatatypeProperty a owl : DatatypeProperty ; 115 rdfs : domain this : PropertyOwner ; 116 rdfs : range xsd : integer . 117 118 this : customTypeDatatypeProperty a owl : DatatypeProperty ; 119 rdfs : domain this : PropertyOwner ; 120 rdfs : range this : DivisibleByFiveEnumeration .
Listing 1.4. Some of the definitions are imported from a separate ontology module. 9 <http :// ontovibe . visualdataweb . org /1.0# > a owl : Ontology ; 10 owl : versionIRI <http :// ontovibe . visualdataweb . org /1.0# > ; 11 owl : imports <http :// ontovibe . visualdataweb . org /1.0/ imported #> ; 12 <http :// purl . org / dc / elements /1.1/ title > " Ontology Visualization Benchmark ( OntoViBe )" .
Listing 1.5. Sets of properties connected to the same classes allow for testing whether a visualization positions such properties in a non-overlapping way. This example shows two cyclic properties connected to the same class. 173 this : cyclicProperty2 a owl : ReflexiveProperty ; 174 rdfs : domain this : MultiPropertyOwner ; 175 rdfs : range this : MultiPropertyOwner . 176 177 this : cyclicProperty3 a owl : ObjectProperty ; 178 rdfs : domain this : MultiPropertyOwner ; 179 rdfs : range this : MultiPropertyOwner .
In ontology visualizations, sets of properties between the same pair of classes (or the same class and literal) pose a particular challenge, as they may lead to overlapping and thus illegible representations. Several of these cases have been considered in OntoViBe. For instance, Listing 1.5 shows two cyclic properties (i.e., properties whose domain and range are identical) connected to the same class.
Finally, a few of the ontology elements are provided with labels, to check how visualizations cope with multilingual labels that may also contain nonASCII characters (Listing 1.6). For all non-ASCII characters, the escaped ASCII representation is used in the ontology file, as that maximizes the chances for a good compatibility with the parser reading the file. 3.2
To verify that OntoViBe covers most of the features defined by the OWL 2
specifications, we provide two coverage matrices. Table 1 juxtaposes the elements of
OntoViBe with systematically listed OWL 2 features as described in the
specifications. Table 2 shows which OntoViBe elements use which concrete OWL 2
identifiers, as per the IRIs declared in the OWL 2 Namespace Document [
The tables also reveal some parts of OWL that are intentionally not included in OntoViBe: Cardinalities: OntoViBe defines only a few cases of cardinality constraints for properties: Structurally, these can be distinguished as no cardinality, cardinality on one end of a property relation and cardinality on both ends of a property relation. Regarding the concrete cardinality constraints applied, exact cardinality, a minimum cardinality, and a combination of a minimum and a maximum cardinality are included in OntoViBe. Moreover, one of the cardinality constraints is qualified and thus applies only to instances of a specific class. These cases can thus only be used for checking whether cardinalities are displayed at all.
We have opted against integrating all supported cardinalities in OntoViBe, as the number of possible combinations would be considerable—in particular, when considering that “special values” such as zero and one might be displayed in special ways. The total number of properties in OntoViBe would have to be increased while providing only minor additional insight into the tested visualization.
Annotations: Informative metadata has no effect on the conceptual structure of an ontology, which is focused in OntoViBe. For that reason, only the most prevalent metadata attributes, such as labels or the ontology title, have been integrated into OntoViBe.
Equivalent constructs: In cases of conceptually equivalent ways to express
statements in OWL, only one way was integrated into OntoViBe. For
instance, deprecation of ontology elements can either be expressed by adding
the owl:deprecated attribute or by declaring the element as belonging to
one of the classes owl:DeprecatedClass or owl:DeprecatedProperty.
Deprecated elements: Deprecated language constructs of OWL itself are not
used in OntoViBe. An example is owl:DataRange that has been deprecated
as of OWL 2 in favor of rdfs:Datatype [
Moreover, statements referring to particular individuals have not been included, as OntoViBe focuses on visualizations of the TBox of ontologies. 4
In the following, we demonstrate the usefulness of OntoViBe by applying four
ontology visualizations to it: SOVA, VOWL, OWLViz, and OntoGraf. The latter
two come with the default installation of the popular ontology editor Prot´eg´e [
We present screenshots of all these ontology visualizations that give an impression of the supported features. We point out peculiarities of the visualization approaches and their implementations that become apparent based on OntoViBe. By this, we would like to provide some examples of how to use OntoViBe for the qualitative analysis of ontology representations, and to confirm that such an analysis is feasible by using OntoViBe.
It should be noted that a comprehensive analysis of ontology visualizations requires additional methods, such as a checklist comprising further evaluation criteria. These methods are typically not generic but tailored to the type of visualization. For instance, measures for graph visualizations of ontologies could include the total number of edges and edge crossings. However, such additional measures are outside the scope of this work. 4.1
OntoViBe:
Ontology
SOVA is a plugin for Prot´eg´e that provides graph visualizations of
ontologies [
Furthermore, some of the limitations of the SOVA implementation can be identified. PropertyOwner and MultiPropertyOwner are classes, but at the same time, they are instances of the class PropertyOwnerType. SOVA displays them twice, once as classes and once as individuals, rather than as a single concept. Cardinality constraints are displayed as small colored circles, which are easy to spot—for a combined minimum and maximum cardinality constraint, however, only the lower bound (the number 5 in the green circle) is shown. Overall, the visualization contains many edges and edge crossings, which significantly reduce its readability. A large number of these edges result from the fact that all implicit subclass relations to owl:Thing are depicted in the SOVA visualization, and that every piece of information is shown in a separate node.
Thing equivalentkkk [dummyProperty] complemenkkk
AlsoEquivalenkkk [SubclassBmEquivalekkk]
PlainClass
DivisibleByTwkkk DivisibleByFivkkk Thing
subproperty
Subclassmof Imp1oerxtetedrnCallAass
classToClkkk Class
Literal disjunktemKlasse 1disjointA
Subclassmof
Intersekkk
Literal 1discjoyicnltAicProcplakksksTcolCaslkskTkoClkkk multiLpropertymkkk cyclicPropkkk cyclicPropkkk date
Literal
Divisikkk
Thing rational hexBinary functionalkkk 1functionalA FunctionalAnchor
UnionDkkk integer 4.2
PropertyOwnekkk DeprecatedImpkkk 1deprecatedBmexternalA
Subclassmof
MultiSubclass
DeprecatedClass 1deprecatedA
OntoViBe visualized with
VOWL. to ous
VOWL, the both obtain attributes
Visual Notation for OWL Ontologies, was
a structural overview of OWL ontologies
of ontology elements at a glance [
been implemented
in two different tools, a plugin for Prot´eg´e and a responsive web application.
Figure 2 has been created with version 0.2.15 of the web application (called
WebVOWL [
Applying VOWL to OntoViBe shows some typical characteristics of VOWL visualizations, such as equivalent classes being represented as one class with a double border, other special elements being multiplied in the visualization (e.g., owl:Thing), or text in brackets below the labels indicating attributes such as functional or symmetric. Like in SOVA, custom data ranges are not (yet) completely shown, as is seen by the respective nodes that simply display the names of the data ranges (e.g., “Divisib...” for “DivisibleByFiveEnumeration”) but no more information on how they are defined. Moreover, it becomes apparent that the WebVOWL implementation tends to route edges between the same pairs of nodes in a way so as to avoid overlapping labels. As the implicit subclass relations to owl:Thing are not shown in VOWL, the nodes of the graph visualization are less connected than in SOVA. 4.3
The OWLViz visualization is aimed at visualizing exclusively the hierarchical class structure of ontologies. When used to visualize OntoViBe (Figure 3), the fact that only inheritance (“is-a”) relationships are shown by OWLViz 4.1.2, gets apparent. Also, it gets clear that equivalence relationships between classes are expressed as bidirectional inheritance. Other property relations are not visualized in OWLViz, and also further class or property characteristics are not included, making OWLViz an ontology visualization with very limited expressiveness. 4.4
OntoGraf (Figure 4) depicts property relations between classes with colored lines. Given that OntoViBe includes properties of different types and with different characteristics, it is notable that these are not displayed by OntoGraf 1.0.1 in an inherently distinct way. Again, the graph is highly connected, as the implicit subclass relations to owl:Thing are explicitly shown.
Like in OWLViz (Section 4.3), equivalence between classes is displayed by two
opposite inheritance arrows. Additionally, classes that are equivalent to others
are highlighted by an equivalence symbol. Other than that, the test shows that
OntoGraf copes well with several cyclic properties applied to the same class.
LODE is a documentation generator for ontologies [
Based on the OWL 2 specifications, we have defined OntoViBe, a benchmark ontology for testing ontology visualizations. We did not focus on scalability or other common benchmark goals, such as execution speed, but rather on feature completeness and flexibility in terms of combination of elements with regard to the OWL 2 specifications. Since OWL may further evolve in the future, OntoViBe needs to keep being updated accordingly.
Features not included in OntoViBe may be considered for future adjuncts of the ontology. For instance, these could be separate modules that focus on testing specific aspects, such as combinations of cardinality constraints or the population of OntoViBe with individuals and other ABox information.
Generally, we hope that our experiences from the development of OntoViBe can benefit other projects, including benchmark data models beyond the task of ontology visualization. g ip ip ip ip ty llseeagbd rrteeeeoppnnwrtteeeooppnwwltsssceeeeannwlttssceeeoanwwrtreoypnp frrseeooppd frrteeooypp lifrsssceoaap lifrsssceoaap illfttsssceoa liftse lrtseopnp lltssceean
re s ty ite
Namespace prefix a: denotes elements of the core ontology, while elements from the additional module are marked with b:. a ittsehwRiIrseovnRiIrseovnn ifonnuOjtrtrteceoobppPOtttrreaoaoyppPDlttreeaaguViiItrteagvnd iIrsceovund lsseeoauVmsseaAmilifreaaqdudCijtrrseooyppDirrteoaypphnCiirrrseoopnVrtreooypnP irtreeoonpP feonOtteoaaynpDtoaaagnnRDlssoanCianuQmirandCmrseebmmlaaxuQmraaxCmirseevnOirttseecon iitcoanpmitrsopmlseaahuVlfseahS seayhKileeavqun ileeavqun iittscndMiijtsodnWiijtsodnnUiffrtreednF rteceeadpd tteaaoydpCmltceeopnOmmiilrtcaaynd jttoobbOmtttooabDmrcaakbdwCirtsseaonP tteaoanndTtteaoanndS tteaoanndP lllrseaauFViitrseavnPTighnTitrceySPmmiirttseconRlifreeeoxvpRltrooogynpOltoogynOjtrtecbPOitoghnNiteegavPNIeaddnNmilfIrreeexvP IrsecevnnuF iltcaonunFP rteceeapdPDrteeceapdCDttreaaoypPDtegaaanRDlssaCioxAmsyAmmtoannAtoannAilljsADllisAD nNoatmeseseplaecmeents of the core prefix a: deo jo ontology, while elements irc I fo tre ito ito in in iton laud liaud roFm in in A irtrecoypP trreoypnP n tP tC farroemmtahrkeeaddwdiittihonba:l. module s y iltay itthWioxm rty y irtrseeop lsssea
P P op rteopP ltssaCrseebmith ifonOom lepmf rtrteecopP trreoaypP ilteaopbmWtrreoyp trega recou rtreop omrreop rreop rtey rtey rey rtreoyp liiavud rtreoyp litoanP trreoyp rrteoyp lssa rtep en y ty ty t O y f