=Paper=
{{Paper
|id=Vol-1963/paper496
|storemode=property
|title=Alignment Cubes: Interactive Visual Exploration and Evaluation of Multiple Ontology
Alignments
|pdfUrl=https://ceur-ws.org/Vol-1963/paper496.pdf
|volume=Vol-1963
|authors=Valentina Ivanova,Benjamin Bach,Emmanuel Pietriga,Patrick Lambrix
|dblpUrl=https://dblp.org/rec/conf/semweb/IvanovaBPL17a
}}
==Alignment Cubes: Interactive Visual Exploration and Evaluation of Multiple Ontology
Alignments==
https://ceur-ws.org/Vol-1963/paper496.pdf
Alignment Cubes:
Interactive Visual Exploration and Evaluation
of Multiple Ontology Alignments
Valentina Ivanova1 , Benjamin Bach2 , Emmanuel Pietriga3 , and Patrick Lambrix1
1
Swedish e-Science Research Centre and Linköping University, Sweden
2
University of Edinburgh, United Kingdom
3
INRIA, LRI (Univ Paris-Sud & CNRS), Université Paris-Saclay, France
Abstract. The quality of ontology alignments is evaluated by comparing to ref-
erence alignments and calculating general measures, i.e., precision, recall and
F-measure. These measures, however, only provide an overall assessment of the
alignments’ quality, but do not reveal differences and commonalities between
alignments at a finer-grained level such as, e.g., regions or individual mappings.
This demands comparative exploration of alignments at different levels of gran-
ularity. Furthermore, reference alignments to compare to are often unavailable,
which makes such comparative evaluation even more important. To address this
issue, we introduce Alignment Cubes which supports efficient interactive visual
exploration of multiple alignments at different granularity levels.
1 Motivation
Ontology matching is an active research area and many tools and approaches have
been developed in the last 15 years. Tools are evaluated by comparing their alignments
against reference alignments - RAs - and computing measures, such as precision, recall
and F-measure. These measures, however, provide only an overall assessment of align-
ments’ quality and cannot reveal in what aspects one tool outperforms another. Further-
more, RAs are often not available and these evaluation measures cannot be computed.
To understand the strengths and weaknesses of a tool and to compare it to others,
we seek answers to questions such as: Which mappings are computed by all or most
tools? Are there mappings which are rarely computed or not at all? Do tools compute
mappings for the same regions of the ontologies? [4]. These questions demand flexible
exploration and comparison of the alignments at different granularity levels and cannot
be answered by the aforementioned measures. Practitioners resort to writing custom
scripts, which can be error-prone, and time- and effort-consuming as such scripts have
to be crafted for every single question. Besides, comprehending their results is cumber-
some especially when the size and number of alignments grow.
In our main conference paper [5], we identify several evaluation use cases and dis-
cuss shared activities that could be efficiently supported through interactive visualiza-
tion. To address these use cases, we present Alignment Cubes—an interactive visualiza-
tion environment for the comparative exploration and evaluation of multiple alignments
at different levels of granularity. This demo paper is a companion paper to this research
Fig. 1: Alignment Cubes User Interface
paper. As this is an interactive tool, it will be best demonstrated in an interactive setting
where attendees can interact with it using their own datasets.
Related Work: Existing evaluation frameworks (SEALS4 , KitAMO [7] and AMC [8])
have focused mostly on back-end features, such as storing alignments, configuring and
executing algorithms. Little attention has been devoted to the interactive exploration of
alignments. Interactive visual approaches have the potential to efficiently support such
interactive exploration by taking benefit from the humans’ powerful visual perception
system. Similarly to one of Alignment Cubes’ small multiples view, AgreementMaker
[2] visualizes several alignments as juxtaposed matrices. Another tool capable of pre-
senting several alignments is VOAR [9]. However, the view quickly becomes cluttered
as the size of ontologies and the number of alignments grow.
2 Alignment Cubes
In [5], we identify several analytic tasks shared by our use cases, that could be effi-
ciently supported by interactive exploration at different granularity levels to perform,
e.g., compare and contrast tasks. These high-level features together with our previous
analysis of user interfaces for ontology alignment [6,3] served as guidelines for the tool
we developed. We drew from state-of-the-art approaches in the field of dynamic net-
work visualization, and identified Matrix Cubes [1] as a promising visual approach to
serve as a foundation for our tool - Alignment Cubes5 .
4
http://seals-project.eu — Semantic Evaluation At Large Scale
5
http://www.ida.liu.se/˜patla00/publications/ISWC17 provides supple-
mental material to this submission: a screencast and a downloadable version of the tool itself.
2.1 Alignments Presentation
To make efficient use of the available screen real-estate, an alignment is presented as a
matrix where the rows hold the concepts from one of the ontologies and the columns
hold those from the other. The ontologies are depicted as expandable and collapsible
indented lists (according to their taxonomic relationships) on both sides of the matrix.
Cells denote existing mappings between concepts in the respective rows and columns.
Stacking several matrices (alignments) on top of each other creates an alignment cube.
Fig. 1 shows the initial view of the tool. Two of the ontologies from the OAEI
Conference track, ekaw (columns, 77 concepts) and confOf (rows, 38 concepts) are
depicted on the red and green axes. 12 alignments are laid out along the blue axis—the
RA for 2016 and alignments computed by the LogMap-family of systems from 2011 to
2015. Each alignment is color-coded to visually differentiate the mappings, by grouping
the cells that belong to each of them using a pre-attentive variable.
2.2 Granularity Levels
To support views at different levels of granularity—from an overall view to regions
based on the is-a hierarchy, and down to single mappings—we introduced alignment
modes. In similarities mode a filled cell represents an existing mapping between a pair
of concepts. In mappings mode, a filled cell indicates that there is at least one existing
mapping between a pair of concepts or their descendants. The cell weight represents
either the similarity value (in the former case), or the number of mappings (in the latter
case). Each mode is focused on performing one of two tasks: to compare similarity
values for a pair of concepts, and to identify regions in the alignments with few or
many mappings. The latter task provides a starting point for exploration and highlights
regions of interest where many or few mappings have been calculated. When a concept
is expanded in mappings mode, a cell is shown for both the concept itself and its sub-
concepts. This forms regions in the cube where smaller cells indicate mappings deeper
in the hierarchy.
2.3 Interactive Visual Exploration
The Alignment Cubes user interface provides a variety of interactions to support visual
exploration, shown in fig. 1: changing alignment modes (see above), cell color and size
encodings, switching between individual views, adapting cell transparency, brushing
and linking, as well as alignment slice reordering. For example, cells can be filtered out
by specifying minimum or maximum value thresholds using a range slider. This also
allows to simulate different thresholds and explore what-if questions and cases. Entire
alignments can be hidden. To support pattern discovery and to facilitate comparison,
the order of alignments (slices) in the cube can be changed based on precision, recall,
F-measure or a matcher name.
2.4 Compare and Contrast
Alignment Cubes provide several views onto the data, resulting from manipulations of
the 3D cube. The individual views are: (a) 3D view, (b) 2D projection on orthogonal
faces of the cube, (c) side-by-side layout of the cube’s slices (juxtaposition)—small
multiples view, (d) in-place rotation of individual slices for quick preview. The 3D cube
provides an overview of the number of alignments, and number, size, and distribution
of cells (mappings). It helps identify regions of interest and thus drive the initial ex-
ploration phase, and can possibly yield some high-level insights (on fig. 1 we quickly
notice that a mapping is present in the RA (the single brown cell) but not in the other
alignments). It allows for interactive rotation and zoom, but suffers from the typical
drawbacks of 3D visualization, including occlusion and perspective distortion. Projec-
tion views allow for a clutter-free aggregated view on all alignments by orthogonally
overlapping cells. Side-by-side views provide the most detailed view onto the data by
entirely decomposing the cube and showing each alignment in detail. Individual views,
together with the ability to vary cell size, color, and translucency, allow for flexible
multi-perspective exploration of the entire data set.
Each of the two projections (alignment topology and concepts network) is paired
with its respective side-by-side view. Both projections/side-by-side pairs allow for in-
vestigating the behavior of matchers—the alignment topology pair focuses on the simi-
larities and differences between the alignments as a whole, while the concepts network
pair allows for analyzing the behavior of matchers for a particular concept.
3 Demonstration Scenario
During the Demo session we will demonstrate the different views and exploration fea-
tures of the tool by conducting a comparative evaluation of several alignments. We
would also like to invite participants to the Demo session to use their own datasets with
Alignment Cubes6 .
References
1. B Bach, E Pietriga, and J-D Fekete. Visualizing dynamic networks with matrix cubes. In CHI,
pages 877–886, 2014.
2. I Cruz, C Stroe, and M Palmonari. Interactive user feedback in ontology matching using
signature vectors. In ICDE, pages 1321–1324, 2012.
3. Z Dragisic, V Ivanova, P Lambrix, D Faria, E Jiménez-Ruiz, and C Pesquita. User validation
in ontology alignment. In ISWC, pages 200–2017. 2016.
4. Z Dragisic, V Ivanova, H Li, and P Lambrix. Experiences from the anatomy track in the
ontology alignment evaluation initiative. 2017. submitted.
5. V Ivanova, B Bach, E Pietriga, and P Lambrix. Alignment cubes: Towards interactive visual
exploration and evaluation of multiple ontology alignments. In ISWC, 2017.
6. V Ivanova, P Lambrix, and J Åberg. Requirements for and evaluation of user support for
large-scale ontology alignment. In ESWC, pages 3–20. 2015.
7. P Lambrix and H Tan. A tool for evaluating ontology alignment strategies. J Data Semantics,
VIII:182–202, 2007.
8. E Peukert, J Eberius, and E Rahm. Amc-a framework for modeling and comparing matching
systems as matching processes. In ICDE, pages 1304–1307, 2011.
9. B Severo, C Trojahn, and R Vieira. A gui for visualising and manipulating multiple ontology
alignments. In ISWC (Posters & Demos), pages 37–48, 2015.
6
If you are interested to do so, please contact the first author in advance.