89
Chapter 4
AlViz: A Tool for
Ontology Alignment
Utilizing Information
Visualization Techniques
Monika Lanzenberger
Vienna University of Technology, Vienna, Austria
Jennifer Sampson
Statoil, Bergen, Norway
Contents
4.1 Introduction to Ontology Alignment ........................................................ 90
4.2 Introduction to Information Visualization .................................................91
4.3 Visualization for Mapping and Alignment .................................................92
4.4 AlViz: Multiple View Visualization for Semi-Automatic
AlignmentofOntologies ............................................................................94
4.5 Ontology Alignment in AlViz ....................................................................99
4.5.1 Alignment Algorithm .....................................................................99
4.5.2 Interpretation of Alignments ........................................................102
4.6 Need for Visualization in Ontology Alignment ........................................105
References .........................................................................................................106
90 ◾ Monika Lanzenberger and Jennifer Sampson
4.1 Introduction to Ontology Alignment
e main purpose of ontology alignment is to determine which entities or expres-
sions in one ontology, correspond to other entities in a second ontology. Like Ehrig
et al. [2005], we dene an ontology as a tuple: O = (C, H
C
, R
C
, H
R
, I, R
I
, A).
Concepts C are organized in a subsumption hierarchy H
C
. Relations R
C
are between
pairs of concepts and may also be arranged in a hierarchy H
R
. e instances of
specic concepts I are interconnected through relational instances R
I
. A represents
the axioms used for inferring knowledge. We dene ontology alignment as given
two ontologies O
1
and O
2
, each describing a set of discrete entities (concepts C,
relations R, and instances I), and nd the relationships (equal, syntactically equal,
broader than, narrower than, similar to, and dierent) that hold between these
entities [Sampson, 2007]. Ontology alignment also includes evaluating the cor-
respondences, sometimes bringing about the need to transform one or more input
ontologies.
e use of terms cannot be expected to be consistent between related but
separate ontologies. Dierent parts of the ontologies may contain conicting or
ambiguous elements for concepts, instances, and relations. However, we can nd
correspondences between ontologies in three main ways. First, intentional meth-
ods compare ontology schemas from terminological and structural perspectives.
Terminological correspondences between entities are based on names, labels, or
descriptions of entities. When comparing labels or names, techniques such as string
equality, string dissimilarity, or edit distance may be used, resulting in what we
dene as a syntactical match.
String equality returns 0 if the strings compared are not the same and 1 if they
are the same. e edit distance between two strings is the minimal number of
changes required to transform one string into another. e structures of entities
in two ontologies can also be compared from internal and external perspectives.
Comparing the internal structures of ontologies involves checking their property
ranges, cardinalities, and the transitivities and/or symmetries to determine simi-
larities of the structures. ese types of comparisons are most often used in con-
junction with other techniques for alignment as they are not good indicators for
similarity when used alone.
External structure comparison involves comparing the hierarchical positions of
the entities in two ontologies. e intuition is that if two entities are similar, then
often their neighbors are similar. Second, extensional methods compare the set of
instances of the classes in the ontologies. is type of comparison can be used when
the classes share the same instances or when they do not share the same instances.
Testing for intersection between two classes is one extension comparison approach.
e technique can be rened through the calculation of the symmetric dierence
between the two extensions. ird, a semantic comparison involves comparing the
interpretations of the entities in the ontologies. Semantic methods have model-
theoretic semantics, that is, they use deductive methods to verify the results. For
AlViz ◾ 91
example, description logics techniques such as the subsumption test can be used
to establish relations between classes during ontology alignment. In the FOAM
[Ehrig, 2005] algorithm, all three types of comparisons are made between enti-
ties. e resulting similarities provide evidence that two entities are the same (or
similar) and can potentially be aligned. Calculating the similarity between two
entities requires a range of similarity functions that combine dierent features of
the ontologies with appropriate similarity measures. Section 4.4 provides a detailed
example of ontology alignment.
4.2 Introduction to Information Visualization
Visualization has appealing potential when it comes to creating, exploring, or
verifying complex and large collections of data such as ontologies. In particular,
Information Visualization (InfoVis), which deals with abstract and non-spatial
data, oers a bundle of techniques to represent hierarchical or semi-structured data.
us it is no surprise that many ontology tools integrated visualization in some
fashion during the past decade. Many tools rely on simple types of visualizations
like two-dimensional trees or graphs. Usually the nodes stand for concepts and the
edges represent relationships of concepts, but other approaches exist as well.
A literature study indicated a broad interpretation of ontology visualization dif-
fering among the various tools. InfoVis uses visual metaphors to ease the interpre-
tation and understanding of multidimensional data to provide users with relevant
information. A visual metaphor consists of graphical primitives such as point, line,
area, or volume and utilizes them to encode information by position in space, size,
shape, orientation, color, texture, and other visual cues, connections and enclosures,
temporal changes, and viewpoint transformations [Card et al., 1999]. e goal of
InfoVis is to promote a more intuitive and deeper level of understanding of the inves-
tigational data and foster new insights into the underlying processes [Tufte, 2001].
An enormous amount of work was done in the eld of InfoVis in recent years. e
methods range from geometric techniques such as scatter plots and parallel coordi-
nates [Inselberg and Dimsdale. 1990]), glyphs like InfoBug [Chuah and Eick, 1997],
icon-based techniques like Cherno faces [Cherno, 1973], stick gures [Pickett and
Grinstein, 1988], pixel-oriented recursive patterns [Keim et al., 1995], and spiral and
axis techniques [Keim, 1996] to interactive visualizations for hierarchical informa-
tion such as cones or cam trees [Robertson et al., 1991], hyperbolic trees [Lamping
et al., 1995], graph-based techniques such as small world graphs [van Ham and van
Wijk, 2004], maps such as themescape [Wise et al., 1995], distortion-oriented meth-
ods like the sheye lens [Furnas, 1986], other focus + context techniques [Pirolli et
al., 2001], and hybrids like Stardinates [Lanzenberger et al., 2003].
Combining several views is well known as multiple view visualization, which
oers several advantages such as improved user performance, discovery of unforeseen
relationships, and desktop unication [North and Shneiderman, 1997]. Generally,
92 ◾ Monika Lanzenberger and Jennifer Sampson
in InfoVis the exploration process is characterized by cognitive abstraction. In addi-
tion, visualization often reduces information or emphasizes certain aspects of the
data in order to ease goal-oriented interpretation. Combining distinct visualizations
yields dierent kinds of abstractions from the data that allow for diverse approaches
of exploration. An important challenge of multiple view visualization is its complex-
ity for the users. ey need to switch between dierent views and contexts.
4.3 Visualization for Mapping and Alignment
A few visualization tools support users with ontology mapping and alignment. We
identied six such tools: OLA, Coma++, PromptViz, CogZ, Optima, and AlViz.
OLA [OWL Lite Alignment; Euzénat et al., 2004], a stand-alone program,
uses graph-based visualizations to represent ontologies. In particular, an extended
JGraph API is applied. e graph structure of OLA makes relationships between
language elements explicit, e.g., if a class c refers to another class c′ via an
owl:allValuesFrom restriction, a labeled path between the corresponding nodes in
the OL graph is shown such that the connection between both classes is perceived
intuitively. Besides common subclass relationships, users can activate a display of
edges between objects that are reverse, symmetric, or transitive.
Coma++ [Aumueller et al., 2005], a stand-alone tool for schema and
ontology matching, uses simple lines to connect mapping pairs in list views.
However, its main focus is the mapping algorithm, not the visualization of
mapping results.
PromptViz [Perrin, 2004] is a visualization tool for Protégé’s Prompt tool [Noy,
2004]. It provides visual representations of the dierences between two versions of
an ontology using histograms within a tree map. e bars in the histograms repre-
sent the percentages of descendants classied as unchanged, added, deleted, moved
from, moved to, and directly changed, respectively. A histogram is divided into
four linked frames: (1) an expandable horizontal tree layout of the ontology show-
ing the dierences; (2) a tree map layout of the ontology embedded in a zoomable
user interface; (3) a path window that shows the location of the currently selected
concepts within the is-a hierarchy of the ontology; and (4) a detailed list of the
changes (if any) to the currently selected concept.
Implemented as a user interface plug-in extension to Prompt, the CogZ tool
[Falconer and Storey, 2007] oers visual mapping functionality. It enables users to
examine, add, or remove temporary or permanent mappings. A bundle of ltering
options help handle the complexity of ontology mapping. Moreover, neighborhood
graphs, sheye lenses, and other tree map or pie chart views oer ecient means for
exploration of the mappings. (see Figures 4.1 and 4.2).
Optima [Kolli and Doshi, 2008] is a Jena-based alignment tool using a graph-
theoretic algorithm to nd the most likely match between two ontologies (opti-
mization) and computes the likelihood using the expectation maximization (EM)
AlViz ◾ 93
technique. It involves structural and lexical similarities between schemas. Both
ontologies are visualized as graphs available with several dierent layouts such as
tree, circle, etc. However, this tool does not integrate other alignment algorithms
and is not linked to an ontology editor.
AlViz [Lanzenberger and Sampson, 2006] is a research prototype for visual
ontology alignment implemented as a multiple-view plug-in for Protégé (see
Figures4.1 and 4.2). Based on similarity measures of an ontology matching algo-
rithm like FOAM [Ehrig, 2005], AlViz helps assess and optimize the alignment
results at dierent levels of detail. Clustered graphs enable users to examine and
manipulate mappings of large ontologies. In AlViz in conjunction with FOAM,
Figure 4.1 AlViz: the four views of the tool visualize two ontologies named tou-
rismA and tourismB. The nodes of the graphs and dots next to the list entries rep-
resent the similarities of the ontologies by color. The sizes of the nodes result from
the number of clustered concepts. The graphs show the IsA relationship among
the concepts. Light gray/green indicates similar concepts available in both ontolo-
gies. Dark gray/red nodes represent equal concepts. The sliders to the right adjust
the level of clustering. (The figure is available in color at: http://www.ifs.tuwien.
ac.at/~mlanzenberger/alviz/graphics/ASWT/Figure41_AlVizScreenShot1.pdf)
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.