Abstract
Visualization is a key technology forunderstanding large
bodies of data. Our approach to visualizing abstract, non-
geometric data involves a reduced-representation
overview, multiple linked views, filtering and focusing
techniques to reduce visual clutter, color, anda highly-
interactive user interface. The reduced representations
allow users to see the entire data set in one view while still
providing immediate accessto relevant detail and answers
to specific questions in the linked views. Wehave
developed a software infrastructure embodying our design
principles for producing novel, high-bandwidth
visualizations of corporate datasets. Our approach to
abstract data visualization is one the best off-ramps on
theinformation superhighway.
Keywords:
Visualization, Graphic Interaction, Abstract
DataVisualization, Database Visualization, Data Mining
Introduction
Just as spreadsheets revolutionized our ability to understand
small amounts of data, visualization will revolutionize the
way we understand large corporate datasets. Our research
focuses on extracting the information latent in large
databases using novel visualizations.
The difficulty in extracting this information is
understanding the complexity of the databases. To aid in
this task, we have created many novel, highly
interactive visualizations of large datasets. Our research
involves developing the techniques, software tools, and
infrastructure to mine knowledge fromcorporate databases
so that it can be put to competitive and
commercial advantage.
INTERACTIVE VISUALIZATION
Most of today's interfaces to large data sets show only a
few aggregate items at a time. Our goal is to use every
available screen pixel to show as much data as possible,
thereby providing local detail in a global context.
To achieve this goal we use interactive techniques to solve
the clutter problems associated with information-dense
displays.
Compact Graphic Representation
Navigation is frequently anissue in the design of interactive
systems dealing with large information spaces. Compact
graphic representations can provide global perspective in
aninformation-rich environment, thereby maximizing data
accessibility and minimizing navigation. These compact
representations take full advantage of perceptual cues (size,
position, color, depth, sound, etc.).
Highly Interactive Linked Views
The power of our representations is magnified through the
use of interaction and linked views Each view, whether
custom or standard (color keys, bar charts, box
plots,histograms, scatter plots, etc.), functions both as a
display and a control panel. Selecting and filtering data in
one view instantly propagates to the otherviews, thereby
providing additional insights. Linking multiple
views interactively provides an integrated visualization far
more powerful than the sum of the individual views.
SYSTEMS
Our systems have been used to successfully analyze and
present software version control information, file systems,
budgets, network traffic patterns, consumer shopping
patterns, relational database integrity constraints, resource
usageon a compute server, etc. The amount of information
that our systems present on a single screen is between
10,000 and 1,000,000 records. The systems we have built
include:
- SeeData relational data
- SeeDiff file system differences
- SeeLib bibliographic databases
- NicheWorks[1] abstractnetworks
- SeeLog time-stamped log reports
- SeeNet[2] linked geographic data
- SeeSlice[3] program slices and codecoverage
- SeeSoft[4] lines of text in files
- SeeSys[5] hierarchical software modules
- SeeTree[6] hierarchical data
Since the needs of each user are unique, our visualizations
are task-oriented. Our most successful visualizations
help frame interesting questions as well as answer them.
Our visualizations:
- Make use of existing data.
In many cases large databases of vital importance to an
organization already exist. Our visualizations
extract meaningful information from this data.
- Focus on real problems with targeted users.
Our research efforts are motivated by business needs
and address real problems.
- Leverage interaction.
The interesting subset of a large body of datais highly
volatile and entirely dependent on the user's task.
Dynamicinteraction allows users to separate the wheat
from the chaff.
- Are information dense.
The interesting subsets of a largedata set can easily
consist of thousands of records. Interesting
questions often require a global view of large numbers
of individual items.
- Focus on understanding and insight.
Resultsare more important than any particular
technique.
EVALUATION
Our research interest is in visualization techniques that scale
to industrial-sized systems. By applying our tools toreal
problems and producing working software, we gain an
increased understanding of how visualization methodology
works in practical situations. This enables us to discover
the fundamental insights and formulate the
guiding principles for effectively extracting information
using visualization. By exercising our tools, we gain
insights into critical issues, which enables to us refine our
design principles and improve our capability for
rapidly producing novel, custom, information-rich displays.
Our systems have enabled users both to 1) gain newinsight
into their data and 2) improve performance of existing data
analysis tasks. For example, a formal evaluation of 40
subjects presented with a budgetanalysis task found a
graphic visualization interface to be 50% faster overallthan
a comparable dynamic outline interface [6].
SOFTWARE AND TECHNOLOGY
Underlying all of our visualizations is a common
infrastructure embodied in a C++ library that handles
interaction, graphics, and data linking. This C++
Visualization Library helps us to:
- Minimize our development time,
- Encapsulate expertise and design principles,
- Build cross-platform systems
(UNIX/X11, Open GL, and PC/Windows), and
- Keep visualization application code small.
CONCLUSION
Visualization is a key technology that canhelp users
understand the complexity in industrial-sized systems. We
havedeveloped many interesting and novel visualizations of
a variety of large andcomplex data sets. We would like to
share and discuss our approach to abstract data visualization
with others working in the area of interactive interfaces
to complex information.
ABOUT THE AUTHORS:
Stephen G. Eick is the Technical Manager of the Data
VisualizationResearch group at AT&T Bell Laboratories.
His educational background includes a B.A. from
Kalamazoo College (1980), M.A. from the University of
Wisconsin at Madison (1981), and a Ph.D. in Statistics from
the University of Minnesota (1985). Eick is an active
researcher, is widely published, and holds severalsoftware
patents. He is particularly interested in visualizing
databases associated with large software projects, networks,
and building high-interaction user interfaces.
Brian S.Johnson is a Member of Technical Staff in the
Data Visualizaion Research groupin the Software and
Systems Research Center at AT&T Bell Laboratories.
Hereceived his B.S. in Computer Science from the
University of Minnesota (1987),and his Ph.D. in Computer
Science from the University of Maryland (1993). Heis
actively involved in using visualization techniques to
understand the abstract information associated with large
corporate data sets.
Acknowledgments
The research presented is a joint effort of the following
people: Jackie M. Antis, David L. Atkins, Thomas J.Ball,
Kenneth C. Cox, Stephen G. Eick, Brian S. Johnson, Paul J.
Lucas, John D.Pyrce, Anselm Spoerri, Joseph L. Steffen,
Graham J. Wills.
DEMO References
- Eick, S.G. andWills, G.J. Navigating Large Networks
with Hierarchies, in Proc. IEEEVisualization ‘93, 1993
- Becker, R.A., Eick, S.G.,Miller, E.O., and Wilks, A.R.
Dynamic Graphical Analysis of Network Data. inProc.
ISI, 1989.
- Ball, T.J. and Eick, S.G. Visualizing Program Slices, in
Proc. IEEE VisualLanguages, 1994.
- Eick, S.G., Steffen, J.L., and Sumner E.E.: Seesoftô- A
Tool forVisualizing Line Oriented Software, IEEE
Transaction on Software Engineering11,18 1992.
- Baker, M.J., and Eick, S.G. Space-Filling
SoftwareVisualization, in Proc. Int. Conf. Software
Engineering, 1993.
- Johnson, B.S. Treemaps: Visualizing Hierarchical and
Categorical Data, Ph.D.Dissertation, University of
Maryland, 1993.
