Using Semantics for Interactive Visual Analysis
of Linked Open Data
Gerwald Tschinkel1 , Eduardo Veas1 , Belgin Mutlu1 and Vedran Sabol1,2
1
Know-Center gtschinkel|eveas|bmutlu|vsabol@know-center.at
2
Graz University of Technology
Abstract. Providing easy to use methods for visual analysis of Linked
Data is often hindered by the complexity of semantic technologies. On
the other hand, semantic information inherent to Linked Data provides
opportunities to support the user in interactively analysing the data. This
paper provides a demonstration of an interactive, Web-based visualisa-
tion tool, the “Vis Wizard”, which makes use of semantics to simplify the
process of setting up visualisations, transforming the data and, most im-
portantly, interactively analysing multiple datasets using brushing and
linking methods.
1 Introduction
An objective of the CODE3 project is to make Linked Data accessible to novice
users by providing easy to use methods for visual data analysis. This is hard to
achieve with current Linked Data tools, which require user’s knowledge of se-
mantic technologies (such as SPARQL). This paper demonstrates how semantic
information can be used to support the interactive analytical process, without
the need for users to understand the complexities of the underlying technology.
Within CODE we use the RDF Data Cube Vocabulary4 for describing statis-
tical datasets. Our “Vis Wizard”5 tool provides an intuitive, easy to use interface
supporting visualisation and interactive analysis of RDF Cubes. In the Vis Wiz-
ard we utilise semantic information from Linked Data to support the user in:
1. Selecting and configuring the visualisations
2. Aggregating datasets
3. Brushing and linking over multiple datasets
This paper illustrates the use of semantic technologies in a visual analytics
tool that enables novice users to perform complex operations and analyses on
Linked Data. The demonstration focuses mainly on step 3, with a screencast of
the demonstration also being available6 .
3
http://code-research.eu
4
http://www.w3.org/TR/vocab-data-cube
5
http://code.know-center.tugraz.at/vis
6
http://youtu.be/aBfuGhgVaxA
II
Related work: A wide range of tools offers functionalities for visualising and
interacting with data, but only a few rely on semantic information to support the
analytical process. Tableau [6] provides a mighty visualisation toolset, however
it does not make use of semantic information for assisting the user. The CubeViz
Framework [5] facilitates visual analytics on RDF Data Cubes, but does not use
semantics for the user interface. CubeViz supports no brushing, no possibility
to compare datasets directly and no automatic selection of visualisations. Cam-
marano et al. [1] introduces a method to automatically analyse data attributes
and map them to visual properties of the visualisation. Even so, this does not
include an automatic selection of visualisation types.
2 The Linked Data Vis Wizard
The underlying thought is to make the user capable of visually analysing data
without knowing about the concept of Linked Data or RDF Data Cubes. How-
ever, the Vis Wizard utilises the available semantic information to support users
in interacting with the data and performing analytical tasks.
Fig. 1. Two RDF Data Cubes are shown in the Vis Wizard. Brushing the 3G coverage
value in the parallel coordinates highlights corresponding countries in the geo-chart.
Scenario: Figure 1 compares two datasets taken from the EU Open Data End-
point7 in the Vis Wizard. The first one, shown in parallel coordinates, represents
the 3G coverage in Europe, as percentage value, per country for each year. The
second dataset, shown in the geo-chart, contains active SIM cards per 100 people
(encoded by colour-grading) for countries in Europe. In the following we use the
Vis Wizard to gain insights into the data and ascertain the datasets correlate.
2.1 Interactive Visual Analysis
Selecting and configuring the visualisation: The first step is to find an
appropriate visual representation for the given dataset. Within the 10 supported
7
http://open-data.europa.eu
III
charts only those are made available which can actively and meaningfully be
used with the provided data. For example, the geo-chart is only available if the
data contains a geographic dimension. After the chart was selected, the user
can adjust the mapping of data onto the visual properties (e.g. axes, colours,
item sizes etc.) of the chart, whereby only suitable mappings are offered. Chart
selection and the data mapping is computed by an algorithm [3] comparing the
semantic information in the RDF Data Cube with the visual capabilities of the
chart, which are described using the Visual Analytics Vocabulary8 .
Aggregation: We provide a dialogue for aggregating the data and creating a
new Data Cube. In the scenario shown in Fig 1 the second dataset was averaged
over the years and visualised over the countries. Using semantics we differentiate
between dimensions and measures and enable validation of the user choices.
For suggesting charts and supporting aggregation we are utilising RDF datatype,
occurrence and persistence.
Brushing and Linking: The idea behind brushing and linking is to combine
different visualisations to overcome the shortcomings of single techniques [2].
Interactive changes made in one visualisation are automatically reflected in the
other ones. Our scenario contains two separate datasets: the first dataset has
the dimensions “country” and “year”, the second dataset has only “country”.
For conventional tools it is hard to provide interaction over different datasets,
because relationships between them are usually not explicitly available. In cases
when columns are labelled using equal strings guessing the relationships may be
possible, but when labels differ, e.g. a dimension in dataset A is called “Country”
while in the dataset B it is called “State”, the relation cannot be established. In
such cases the burden of understanding the structure of the datasets and linking
them together falls on the user. Within RDF Data Cubes, each dimension has an
URI which is (by definition) unique and can be used to establish the connection
between datasets, making linking and brushing over different datasets possible.
Applied to our scenario the following interactive analysis is performed (see
Fig. 1): The user applies a brush on the first dataset by selecting a specific
value range in the “3G coverage” dimension using the parallel coordinates chart.
Countries outside of the selected range are greyed out in the geo-chart, which
shows the second dataset (SIM card penetration). Obviously, a high 3G coverage
correlates with high SIM card penetration (red), with one exception - France.
It should be noted that the functionality of linking data over different datasets,
or even different endpoints, depends on the quality of the semantic information:
the URIs of the cube-dimensions in different datasets need to be consistent. If
datasets use different, domain-specific URI namespaces, linking the data will not
be possible.
3 Evaluation
We conducted a formative evaluation to explore if our goals regarding the us-
ability of the Vis Wizard could be achieved and to ascertain that users were able
8
http://code.know-center.tugraz.at/static/ontology/visual-analytics.owl
IV
to analyse complex datasets. Eight test users participated which executed six
tasks, where one task was exclusively about linking and brushing. Test users had
a good knowledge of computers, but were not familiar with semantic data. We
conducted a quantitative subjective workload test, using the simplified NASA
R-TLX, and a qualitative thinking aloud test. More details on the evaluation,
including methodology, test users and results are available under [4].
The functionality supporting the choice and configuration of the visualisa-
tion was much appreciated, but users pointed out that immediately suggesting
the most suitable visualisation would have been even more helpful. The task
regarding brushing in the scatterplot had a very high subjective performance of
accomplishing (the median was 91.25 on a scale from 0 to 100, 100 being the
highest value achievable). The conclusion of our evaluation is that, while several
usability issues still need to be fixed, the overall advantage is clearly observable.
4 Conclusion and Future Work
Within this research we have observed a high potential in using semantic infor-
mation for improving interaction in visual analytics. It has been shown that the
user supporting techniques were helpful in gaining insights from the data, with-
out spending much time in selecting and configuring visualisations or analysing
how to link the datasets manually.
As for our purpose the correctness of the semantic annotations of the data is
essential, the stability of our approach could be improved by implementing the
use of URI aliases. We will also explore the possibilities to rank the visualisations
in order to, given a particular dataset, automatically show the most suitable one.
Acknowledgements. This work is funded by the EC FP7 projects CODE (grant 296150) and EEX-
CESS (grant 600601). The Know-Center GmbH is funded by Austrian Federal Government within
the Austrian COMET Program, managed by the Austrian Research Promotion Agency (FFG).
References
1. Cammarano, M., Dong, X.L., Chan, B., Klingner, J., Talbot, J., Halevey, A., Han-
rahan, P.: Visualization of heterogeneous data. In: IEEE Information Visualization
2. Keim, D.A.: Information visualization and visual data mining. In: IEEE Transac-
tions on Visualization and computer graphics (2002)
3. Mutlu, B., Höfler, P., Tschinkel, G., Veas, E.E., Sabol, V., Stegmaier, F., Granitzer,
M.: Suggesting visualisations for published data. In: Proceedings of IVAPP 2014.
pp. 267–275 (2014)
4. Sabol, V., Tschinkel, G., Veas, E., Hoefler, P., Mutlu, B., Granitzer, M.: Discovery
and visual analysis of linked data for humans. In: Accepted for publication at the
13th International Semantic Web Conference (2014)
5. Salas, P.E., Martin, M., Mota, F.M.D., Breitman, K., Auer, S., Casanova, M.A.:
Publishing statistical data on the web. In: Proceedings of 6th International IEEE
Conference on Semantic Computing. IEEE 2012, IEEE (2012)
6. Stolte, C., Hanrahan, P.: Polaris: A system for query, analysis and visualization
of multi-dimensional relational databases. IEEE Transactions on Visualization and
Computer Graphics 8, 52–65 (2002)