KDD Nuggets 96:38, e-mailed 96-12-06
Each tool was designed to help end users drill in to corporate
databases to collect data and view it from the various angles needed
to answer business questions.
Expect a market shakeout in the next two years in which the winners
swallow up niche products to offer versatile,multipurpose tool sets
that support an array of decision-support operations. A key
differentiator among tools will be their ability to support
high-performance, interactive queries across the World Wide Web.
see http://www.computerworld.com/guide/961202drill.html
for full text
cyber23
SQL2VRML - entry space
There is allot of talk in the VR community about the concept
of "database mining", in which a user might extract
information from the database and then "surf" through
it as one might navigate through the WWW. The power of this
concept is extended when the tables that are returned in the
query create visual representations using the scene description
language VRML.
Netscape - Table Index Page
WebSpace - Table Index Space
This particular example uses a database called
"PeopleDB" which contains employee, mailstop, and
department info and therefore contains little scalar data. Even
with one or two scalar fields, visualization can help the user
understand the request by representing the data in a topological
and morphological form. The data becomes a landscape where
differences between indexed objects become obvious quickly.
The advent of dynamic database visualization could assist in
understanding information extracted from queries, and has great
potential in the fields of chemistry, medicine, statistics,
mathematics, finance, investment, real-estate and almost any area
of study where extraction produces large datasets.
PeopleDB - Primary Query Page
One of the goals of this exercise was to rely on an existing
infrastructure of market products to create an information
visualization system. This example is using a Sybase10 SQL server
working in conjuction with Netsite with a SyPerl client which is
making queries and building the VRML worlds. There are two
positive results from this architecture. The first is that any
user that has the Netscape and Webspace products can access this
application at any point of development because everything is
executed on the server side. The second is that the data is
represented "on the fly" so the tables and
visualizations are always created from direct queries at the time
of viewing.
SQL2VRML Query:
select deptname,respext,deptnum from dept where division =
'ADMIN' order by deptnum
Here the user has queried the database to count the number
records grouped by a particular field. The representation of that
query is a table of columns extracted from the query and a 3d
representation of that table. Each object is an abstraction of
what one record has returned from the query.
Index Attributes - Visualizing a Record
Table representations are created using data directly
extracted from tables, that data creates a semantic map to
information that is specified within the query. Information
extracted changes the shape, size and color of the object that
represents its dataset.
Any object within a particular dataset will have identical
index attributes. A consistency of indexes to size, shape or
color will give the user a reference to what data is semantically
effecting the visualization. It should be the case that the field
that effects size for instance, will effect size throughout the
dataset.
When the user makes a primary request to the table based on a
field, each result itself becomes a query of that subset. This is
achieved by writing perl scripts that pass along the required
data from the previous script to write a new one. This new script
executes the new query, taking the user to the next subset of
information, in this way the user "surfs" through the
queries.
Self replicating scripts pass attributes along to their
children without destroying their parents, allowing queries to
grow in granularity by attributes of the query passed by it
parents.
The architectural metaphor serves as a natural point of
reference from which the user can read the data. The
"space" articulates the domain in which a particular
query has taken place. All data extracted from the query resides
within the architectural domain. The space also infers scale and
speaks to the spatial domain of optimum interaction. In this way
the archetypal elements of wall, floor, datum and column guides
the user through the data intuitively without forcing with a
modal interface.
The architectural domain is created from spatial archetypes
such as floor, roof, wall and datum, and provide a spatial point
of reference that distinguishes the data from the set it resides
within. By creating a space the user intuitively knows that no
data within the dataset resides outside of the space.
height='164'>
Dataset Domain - Architectural Metaphor
Wall
The primary bounding object but it also defines openness and
closure of the space. At human scale primary interaction occurs
within the bounding of this element.
Floor
Secondary bounding object defining difference between the user
and sky. Floor is usually the translation point for primary field
data where the primary index field translates the data on the Z
axis.
Datum
Also serves as a bounding object without the breakage of wall,
but the primary intent of datum is to define the scale of primary
interaction.
Column
The primary vertical element. Column is very good at describing
the scale of interaction, while doing very little to define the
domain of the dataset.
Horizon is inherited in the concept of perspective, and with a
horizon line the even small differences in a dataset are read
easily. Much as one might see a small ship on the ocean, in a sea
of data differences give the user visual landmarks to navigate
to.
Visualizing the Dataset - Morphology and Topology
The topology of the dataset is the criteria that defines the
domain of a particular query, it is the rules that make the
landscape. The morphology of the dataset is the criteria that
defines the representation of the objects within that dataset, it
is the rules that make the objects. For instance the data that
defines the parameters of a query would create the topology or
landscape, in which a dataset would be returned, whereas the data
that is extracted from that query would create the morphology or
objects on the landscape.
This demonstration was created using "off-the-shelf"
products using infrastructure that is common in today's corporate
environment. The accessibility of the tools and products required
to create convincing visualizations of SQL databases exists
today. The primary issue at hand is how the data`s representation
is designed in a way that creates "meta-information" or
information that is gained about the information itself. This is
the greatest potential of "database mining", that we
may learn something about the information itself, a whole that
becomes greater than its parts. Without good design and real
consideration about the interaction issues, database
visualization will be little more that a 3d table.
Clay Graham - cyber23@best.com