Multi-view Ontology Visualization
Julia Dmitrieva, Fons J. Verbeek
Imaging & BioInformatics, LIACS, Universiteit Leiden, The Netherlands
Abstract
Available visualization tools are mostly devoted to representation of an ontology only in one certain geometry
and in one chosen dimension. We present our ontology visualization approach where the reasoner is used to
transform an ontology to the graph structure. This graph structure can be visualized with 5 different geometrical
representations, with different level of depth, and with a subset of properties of the user choice. Augmented with
navigation functionalities our visualization becomes a dynamic ontology browser.
1 Introduction
With ontologies researchers can represent, share and integrate knowledge. Currently, there are different tools
available for modeling and editing ontologies, e.g. Protégé[6] and OntoEdit[3]. Besides modeling and editing
also visualization methods are needed, where a knowledge base can be inspected at different levels of granularity
by means of interaction and navigation. From the most comprehensive analysis[8] we conclude that there are no
visualization tools available that can combine 2D and 3D dimensional views, and which can explore an ontology
in different geometries. In each tool the visualization is realized in one particular geometry and in one particular
dimension. With our work we introduce the ontology visualization approach where an ontology can be represented
as a graph structure and visualized by means of different geometrical models.
2 Graph Generation
Ontologies can be very diverse, some are very straightforward, and only represent the taxonomy. Other ontologies
are very complex and based on rich constructors of Description Logic [7]. In NCI-THESAURUS ontology concepts
are frequently defined as union of intersections of other concepts, e.g. C v Au((9R1:B1uB2)t(C1u9R2:C2)).
Because of the diversity and complexity of constructors, we need to find a way to represent both the taxonomy
as well as other kinds of connections between the concepts. In our approach we represent an ontology as a graph
structure. We use a reasoner services to generate a taxonomy and infer another relationships between concepts.
The transformation to graph structure happens as follows:
1. The reasoner finds all subclasses of the given concept C: If B v C then C is connected to B with a red
edge.
2. The reasoner finds all superclasses of the given concept C: If C v B then C is connected to B with a green
edge.
3. Inferring that the C and E are related to each other via R requires extraction of properties from axioms of
the concept. Then the reasoner can be asked whether the concept C u :9R:E is satisfiable, if not then it is
necessary for the class C to have the property R with the filler E.
This procedure begins with the root concept and recursively calls itself till the certain level of depth is reached.
This level of depth is a parameter of the user interface.
3 Visualization
The visualized graph data-structure is directly generated from the ontology (OWL/OBO file) by means of the
transformation process. The graph can be generated for different levels of depth. Level ”1” means that only direct
1

sub/super classes and the concepts that are directly related to the root concept are represented. In level ”2” the
children concepts of the level ”1” concepts are generated. In the visualization a user can navigate through the
graph and expand each concept. In addition, a simple query interface is provided, where the user can retrieve the
concepts which are related to the given query term. With a right-mouse click the definition of a concept can be
retrieved; the definition is an annotation property and has no semantic meaning, it contains the specific domain
knowledge and can be of interest to the specialist.
3.1 Representations of Views Based on Different Relations
In knowledge bases the concepts are interconnected to each other with different kinds of relations. The visualiza-
tion of the graph with all possible relations can be too complicated for the user; consequently, the complex web of
edges may hide important information. Moreover, the user can be interested only in a particular part of ontology
where the concepts are related via a subset of the properties. This representation is referred as a view. The main
idea of the view generation was based on the work of Noy and Musen[11], in which a Traversal View was defined
as a self-contained portion of ontology. We elaborate on this idea for our visualization, with that difference that
in place of extraction of self-contained portion of an ontology we visualize a graph structure that represents this
part of the ontology (cf. Figure 1). In the current version of our application we provide only the global level of
traversal depth, this means that all the properties will be traversed with the same depth. From the list of properties,
the user can select a subset of interest. On basis of this selection the graph structure will be generated only for a
subset of the ontology. This subset contains the concepts that are reachable from the central concept by the defined
set of properties and the defined depth of the traversal.
(a) Visualization of the concept ”quality” from Dolce-Lite
ontology with all relations
(b) The concept ”quality” with all properties and with the
level of depth = 2
Figure 1: Visualization of the concept ”quality” from Dolce-Lite ontology
3.2 Representations of Views Based on Different Geometry
We have implemented two Euclidean views; i.e. Sphere and Disk. In the Sphere visualization, the root concept
is placed in the center of the visualization window. The sub/super classes and related concepts surround the root
concept and are evenly spaced on the imaginary sphere surface (cf. Figure 3). Two other representations, i.e. the
Klein Model, and the Poincaré Disk model (cf. Figure 2), are based on the hyperbolic geometry and realize the
so called ”focus + context” techniques[9, 10]. In addition, we have also implemented the Stereographic view (cf.
Figure 3), where the graph structure is laid out at the surface of a sphere. All the views are augmented with the cor-
responding geometrical transformations. In Euclidean view the Euclidean transforms are implemented, whereas
for the Klein model the hyperbolic transforms[12] are applied, and in Poincaré model the Möbius transformations
are used.
2

(a) Visualization of Pathway ontology with Klein Model (b) Visualization of Pathway ontology with Poincaré
Model
Figure 2: PATHWAY concept with the hyperbolic geometry
(a) Visualization of Pathway ontology with Sphere Model (b) Visualization of Pathway ontology with Stereographic
Model
Figure 3: PATHWAY concept with different geometries
4 Implementation
The implementation was realized in Java. To generate the virtual graph structure, on basis of defined depth
and properties, we have used two additional API’s. The OWL-API [4] is used to parse the ontology and, more
specifically, to extract the concept definitions containing the restriction axioms. The Pellet reasoner [5] is used to
extract the inferred hierarchy. All visualization components are based on the Java-3D API [1].
5 Conclusions and Discussion
In this paper we have presented our approach on information visualization from ontologies. In our visualization
method we represent an ontology as a graph structure. This graph structure is based on inferred hierarchy that is
calculated by the reasoner.
Because the visualization of all the connections between the concepts can be confusing, we provide the user
the possibility to filter the properties and get the requested part of the ontology. In order to explore an ontology,
different geometrical representations can be generated. This multi-view approach can be helpful for the user,
3

because by different geometrical representation an ontology can be shown with emphasizing of different part
of the structure. For example, the hyperbolic views are suitable to represent a global structure of an ontology,
while the Euclidean views are good to visualize a local structure. Our visualization methodology[2] is on-line for
probing and inspection as a Java Web Start application.
At present time we are working on transformation of our application to Protégé [6] plugin.
References
[1] Java 3d api. http://java.sun.com/products/java-media/3D/.
[2] Multi-view ontology visualization. http://www.liacs.nl/~jdmitrie/ontVis/
OntologyVisualization.html.
[3] Ontoedit an ontology engineering environment. http://www.ontoknowledge.org/tools/
ontoedit.shtml.
[4] The owl-api. http://owlapi.sourceforge.net/index.html.
[5] Pellet. http://pellet.owldl.com/.
[6] Protégé, ontology editor and knowledge-base framework. http://protege.stanford.edu/.
[7] The Description Logic Handbook. Cambridge University Press, Cambridge, 2002.
[8] A. Katifori and C. Halatsis. Ontology visualization methods - a survey. AMC Computing Surveys, 39(4),
2007.
[9] J. Lamping and R. Rao. The hyperbolic browser: A focus + context technique for visualizing large hierar-
chies. Journal of Visual Languages and Computing, 7:33–35, 1996.
[10] T. Munzner. H3: laying out large directed graphs in 3d hyperbolic space. Information Visualization, IEEE
Symposium on, 1997.
[11] N. F. Noy and M. A. Musen. Specifying ontology views by traversal. LNCS, 3298:713–725, 2004.
[12] M. Phillips and C. Gunn. Visualizing hyperbolic space: Unusual uses of 4 4 matrices. 1991.
4