The term "Visual Analytics" was
introduced by Jim Thomas [1] [2]. He defined visual analytics as solving data
analysis problems using a facilitating interactive visual interface. The same
term is used to name the scientific discipline about this activity as well.
Currently, visual analytics is widely used
in various areas of human activity — in scientific research, design,
management, financial monitoring [3], information security [4]
and other areas. The key for wide and successful use of visual analytics in
various fields and circumstances — is the well-known predisposition of people
to spatial thinking and image recognition. [5]
Before reviewing Visual Analytics, a few
words about the problem of data analysis in general. We define the data
analysis problem as follows:
•
There is an object of consideration 
.
•
The analyst knows some initial statements about
the object . We call it the “initial
data” 
and it may be presented in one way or
another.
•
It is required to obtain new statements of
interest for the analytic or check the given statements.
In the definition above, the object of
consideration () can be
one of the following:
•
Either one or several real material objects 
.
•
Either one or more imaginary material objects 
.
•
Either one or more abstract objects 
.
As follows from the above definition, visual
analytics is a very broad conception and today we can talk about various forms
of visual analytics. From our point of view, one of the most common forms of
visual analytics is the solution of problems of analysis of various data using the
visualization method.
The solution of the
problem of the initial data analysis consists of the sequential solution of the
following two problems (Fig. 1): [6]
Fig. 1. Visualization method of data analysis
The first problem is the problem of obtaining
a representation of the analyzed data as a graphic image (The problem of
original data visualization). The problem is solved using the computational
power of a computer. The second problem (the problem of obtaining information
of interest to analytics), which is nonetheless important, is the visual
analysis of the obtained image. The analyst has to work on this problem
directly, it cannot be automatized.
Solution of this problem involves a
well-known sequence of steps. The algorithm of translating the initial data to
a graphic image is known as visualization pipeline [7]. The steps are:
•
Sourcing is the
step of obtaining raw data 
. This process may involve combining data from different sources or
just formalizing the data into a computer representation.
•
Filtering is the
step of preliminary processing the data, cleaning it and performing different
calculation tasks to obtain refined data 
. This step is not mandatory and may be voided in some cases.
•
Mapping is the
step of obtaining of the mathematical model of spatial scene 
from the filtered data . Spatial scene is a set of spatial (usually two or three
dimensional) geometric objects 
with corresponding graphical attributes 
. This step is often unique for every data analysis problem. It
determines the effectiveness of the visualization method.
•
Rendering — At this stage, based on the
mathematical model of the spatial scene, its projection graphic image is built 
. The effectiveness of rendering is usually defined by the quality
of the previous step.
A simple pipeline is shown on fig. 2. Data
from an spreadsheet file is converted to the memory as a table, being filtered
and then mapped to equations using linear interpolation. The equation is being
rendered as a regular Y(X) chart.
Fig.
2. Simple visualization pipeline example.
After the completion of the first problem,
the initiative moves to the analyst, one needs to review the graphic images and
do the statements.
As stated above, this problem is solved
directly by the analyst. The results may and will vary depending on the
analyst, their background and goals [8]. The solution of this problem consists
of two generic steps:
•
Visual analysis of the graphic representation. Fundamentally
important to understand that the resulting graphics are only natural and
convenient means of presenting spatial interpretation of the original data for
the analyst. The spatial scene has to be visually analyzed. It allows the
analyst to use the enormous potential possibilities of the spatial thinking in
the analysis process. On this step, the analyst looks on the representation and
notices some regularities or irregularities. This step cannot be strictly
formalized, but we can define the main directions of analysis. The analyst may
look on the following criteria’s:
o
The shapes of the objects on the spatial scene.
An example of this kind of statements is “The object number 1 is a cube, while
object number 2 is a sphere”.
o
The relative positioning of the objects. There
are two basic ways to do these kinds of statements: 1) “objects one to six are
close to each other” or “Object seven is far from all the other objects”. 2) “objects
one to eight are located on the sides of a cube”
o
The optical parameters (e.g. color, opacity
etc.). An example is “Object one is red” or “Object two is opaque”.
Statements
about the visual representation and the objects on the spatial scene are the
results of this step. These statements have nothing to do with the initial data
yet. It is important to highlight that the results of this step will differ
depending on the analyst and their ability to spatial thinking, noticing
details, focus etc.
•
Interpretation of the results with respects to
the initial data. On this step, the analyst converts the statements about the
spatial scene to the object of consideration . This step is one of the most crucial for the method, since it
gives the overall results and it cannot be generally formalized. The analyst
has to understand the connection and transformation of the initial data to the
spatial scene and backtrack it to make the conclusions. The effectiveness of
this step heavily depends on analyst’s understanding of the domain and their
experience. At the same time, this step gives the analyst much deeper
understanding and, according to J. Thomas, insight into the data and domain as
a whole.
An example of a simple visual analysis of
the graphic representation is shown on fig. 3. The chart is being analyzed and
a number of statements regarding the chart are given. Then the data is
interpreted regarding the initial data (from fig. 2).
Fig. 3. Simple visual analysis explanation
Since the interaction between the human and
the computer is the key in the solution of this problem, the solution makes the
visualization method interactive.
After the analysis is complete, the results
may satisfy the analyst, then the analysis of the initial data is complete or
the analyst can decide to go back to one of the previous steps of the problems.
The analyst can decide to repeat either step of either problem. This way, the
method becomes iterative.
Defining the nature of the Visualization
method, one can say that this method is a method of spatial modeling of the
object of consideration. During the solution of the data analysis problem with
the Visualization method, one models the object of consideration as a spatial
scene, makes statements about the scene and then interprets the statements
relating to the object of consideration. Summing up, we can say that the method
is a modeling method with a spatial modeling of the object of consideration . A spatial model is generated based on the object and the results
of the spatial scene analysis are interpreted back to the object
The purpose of the laboratory foundation
was assistance to the physics departments in their studies using the visual
analytics approach. This collaboration designed and implemented a number of
passive and interactive visualization applications for various physics data. Some
of them were developed with cooperation with British National Center of
Computer Animation of Bournemouth University. Invisible objects, for example
small as nanostructures or not physically visible as scalar or vector fields,
got particularly high attention. The developed applications were static or
animated, interactive or passive depending on the goals of the research and the
needs of the researchers.
The focus in physics studies is to
illustrate the processes, so the researcher can see it. As Albert Einstein
said, “If I can't picture it, I can't understand it”. The laboratory main focus
was on making invisible things transparent and easy to observe.
Creating visualizations of nanostructure
models is an example of such transformation. Replacing the pure numbers and
formulas into an animated picture is one of the best ways to help the
scientists understand the underlying processes. Giving the ability to change
the initial data (like the conditions or the amount of initial objects) gives
the analyst the ability to solve their tasks without showing the calculation
results in numbers and wasting time on understanding what went wrong. All the
needed information and processes are transparent and the calculations are done
in the background.
The concatenation of two fullerenes is a
good example of such problem. The created software had switches to tackle the
fullerenes from different angles and with different speed. The analyst could
change the parameters to get the picture they want to see. Two videos are shown
on fig. 4. one is successful concatenation and the second one is not. [9]
Fig.
4. Analysis of collision of two fullerenes
The other highly important object for
visualization would be different fields. Our laboratory in collaboration with
the physics departments worked with both scalar fields and vector (tensor)
fields. In all cases, the main question was about the regions with low or high
values, finding equivalent zones. In these researches, we used both shapes and
graphical attributes to show the values and the directions of the fields.
An analysis of the superconductor parameter
field. The superconductor is modeled according to Ginzburg-Landau theory. The
designed software shows the current lines and the colors depends on the values
of the field. This application allowed the physicists to actually see and
understand the field, the directions and the values of the field.
Fig.
5. Graphical representation of current lines of a vector field of the order
parameter of a superconductor of the second kind
In some cases, there was a need of a
combination of the two objects: a nanoobject and its field. In these cases the
challenge was visualizations of both components in such a way, that they don’t
block each other and one can visually analyze both at the same time.
One of the projects of the laboratory was a
visualization of a nanoobject and the electron density field. Our application
first visualized the nanoobject (as spheres) and then added semitransparent
isosurfaces to show the field. The color of the surfaces depends on the value
of the field. This kind of visualization allows the researcher to see both the
structure and the density of the field around it.
Fig.
6. Visualization of a nanostructure Cl2O and its electron density.
Our laboratory participated in credit
institutes data analysis project managed by the finance department and the Federal
Financial Monitoring Service of the Russian Federation. The main goal was
finding the credit organizations that behave suspiciously, and passing the
information for manual verification and a deeper check. Formalizing the goal,
we can say that it was the anomaly detection problem. Out of all objects of
consideration, the outstanding objects are objects of interest. [10] [11]
A multidimensional visual analysis tool was
created to solve this problem. The main idea behind the tool is doing
additional constructions in the multidimensional space. If the distance in
between two objects in the multidimensional space is less than given by the
analyst measure d (
), then we construct a section in the multidimensional space. The
next step is projection to the 3-dimensional space and mapping the objects to
spheres and the sections to cylinders. We use colors of the cylinders to show
the distance between the multidimensional objects.
Fig.
7. Anomaly detection using the visualization method.
Computational centers generate a lot of
multidimensional metadata about its work. A huge issue is analyzing this data,
since most of it reports normal results, but the interesting results are buried
under huge amount of useless data. Our laboratory worked with a group from CERN
on this project. The main goals were finding which computational centers have
issues. [12]
In order to do this, two applications were
developed — one took the datacenter’s metadata, clustered the data using some
autonomous clustering algorithms and projected the data to a three dimensional
space. The clusters were shown using colors.
Fig.
8. Clustered data (more than 8000 objects with 28 parameters clustered to five
clusters using K means algorithm)
The second application worked with the
networking data and was designed to find the issues in communication between
data centers. To solve this problem, the datacenters were put on two axes and
the data was visualized as a data grid. With this kind of visualization, the
analytic is looking for the spikes on the grid. An example is shown on fig. 9.
The spikes are the red and yellow dots.
Fig.
9. Data grid with the spikes.
Our laboratory does a number of activities
outside of research, but related to visual analysis. The first one is education
on visual analysis and visualization. Currently two courses are being taught
based on the laboratory: a basic one (visual analytics) and an advanced one
(scientific visualization). The first one is aimed at master’s degree students,
while the advanced one is for PhD students. [6]
The second remarkable activity is the
production of the scientific visualization journal. It is an open access
electronic journal with editorial boards from all over the world. Currently
there are five issues a year, in cooperation with two conferences. Issuing
around ten to twelve articles per issue. The journal is being indexed in a
number of databases: SCOPUS, RSCI and Compendex. Visual analysis is the main
theme of the journal: all the articles are related in one way or another either
to visual analysis or to computer graphics. [13]
In recent years, the activities of our
laboratory are mainly focused on theory and development of applications for
visual analysis of multidimensional data. [14] The research we perform may or
may not use the visualization method, but we always focus on combining
calculations with the interactive visualization. Using this technique, the
analytic gets the best of the both worlds and can use their formally logical
thinking as well as spatial thinking to solve complex problems.
|
[1]
|
J. Thomas, K. Cook, V. Crow, B.
Hetzler, R. May, D. McQuerry, R. McVeety, N. Miller, G. Nakamura, L. Nowell
and P. Whitney, "Human—Computer Interaction with Global Information
Spaces—Beyond Data Mining," Digital Media: The Future, pp. 32-46,
2000.
|
|
[2]
|
J. Thomas and K. Cook, Illuminating
the Path: Research and Development Agenda for Visual Analytics, IEEE-Press,
2005.
|
|
[3]
|
M. L. Huang, J. Liang and Q. V.
Nguyen, "A Visualization Approach for Frauds Detection in Financial
Market," in 2009 13th International Conference Information
Visualisation, Barcelona, Spain, 2009.
|
|
[4]
|
A. A. Cárdenas, P. K.
Manadhata and S. P. Rajan, "Big Data Analytics for Security," IEEE
Security & Privacy, vol. 11, no. 6, pp. 74-76, 2013.
|
|
[5]
|
M. S. Khine, "Spatial
Cognition: Key to STEM Success," Visual-spatial Ability in STEM
Education, pp. 3-8, 2017.
|
|
[6]
|
V. Pilyugin, "Scientific
Visualization Laboratory of NRNU MEPhI," NRNU MEPhI, [Online].
Available: http://sv-journal.org/unl/. [Accessed 25 11 2019].
|
|
[7]
|
V. Pilyugin, E. Malikova, A. Pasko
and V. Adzhiev, "Scientific Visualization As Method Of Scientific Data
Analysis," Scientific Visualization, vol. 4, no. 4, pp. 56-70,
2012.
|
|
[8]
|
D. Keim, F. Mansmann, J.
Schneidewind, J. Thomas and H. Ziegler, "Visual Analytics: Scope and
Challenges," Visual Data Mining. Lecture Notes in Computer Science, vol.
4404, 2008.
|
|
[9]
|
M. Strikhanov, N. Degtyarenko, V.
Pilyugin, E. Malikova, M. Matveeva, V. Adzhiev and A. Pasko, "Computer
Visualization Of Nanostructures Experience At NRNU "MEPHI"," Scientific
Visualization, vol. 1, no. 1, pp. 1-18, 2009.
|
|
[10]
|
I. Milman, A. Pakhomov, V.
Pilyugin, E. Pisarchik, A. Stepanov, Y. Beketnova, A. Denisenko and Y. Fomin,
"Data Analysis Of Credit Organizations By Means Of Interactive Visual
Analysis Of Multidimensional Data," Scientific Visualization, vol.
7, no. 1, pp. 45-64, 2015.
|
|
[11]
|
I. Milman and V. V. Pilyugin,
"Interactive Visual Analysis of Multidimensional Geometric Data.,"
in 24 th International Conference in Central Europe on Computer Graphics,
Visualization and Computer Vision WSCG 2016, Plzen, Czech, 2016.
|
|
[12]
|
T. Galkin, M. Grigoryeva, A.
Klimentov, T. Korchuganova, I. Milman, S. Padolski, V. Pilyugin, D. Popov and
M. Titov, "The new approach to monitor the workflow management system
ProdSys2/PanDA of the ATLAS experiment at LHC by using methods and techniques
of visual analytics," Scientific Visualization, vol. 10, no. 1,
pp. 77-88, 2018.
|
|
[13]
|
"Scientific
Visualization," [Online]. Available: http://sv-journal.org/. [Accessed
25 11 2019].
|
|
[14]
|
O. Maslennikov, I. Milman, A.
Safiulin, A. Bondarev, S. Nizametdinov and V. Pilyugin, "Development Of
A System For Analyzing Of Multidimensional Data," Scientific
Visualization, vol. 6, no. 4, pp. 30-49, 2014.
|
