Transcript VIQING: Visual Interactive QueryING

VIQING
Visual Interactive QueryING
Chris Olston
UC Berkeley
Authors
Chris Olston, Michael Stonebraker,
Alexander Aiken, Joseph M. Hellerstein
14th IEEE Symposium on
Visual Languages
Halifax, Nova Scotia, Canada
September 1st - 4th, 1998
Outline
• Introduction
– Related Work
– Background
•
•
•
•
Visual query results
Specifying visual queries
How VIQING generalizes other work
Status and future work
VIQING
Chris Olston, UC Berkeley
Introduction
• Databases are hard to use
– Difficult to understand data in textual form
– SQL query language hard to learn
• Visual Programming Can Help!
– Database visualization systems (like
DataSplash) display data in graphical form
– VIQING provides a simple interface for
expressing queries over visualizations
VIQING Chris Olston, UC Berkeley
?
Related Work
• Other interfaces offer visual programming
– Visualization
• QBE, Cupid, Tioga-1, AVS, Khoros, MS-Access, DEVise
– Querying
?
• 4GLs, Tioga-1, AVS, Khoros, Access, DEVise, Magic Lenses
• But only VIQING/DataSplash offers a
unified visual programming model for
visualization and visual querying
VIQING
Chris Olston, UC Berkeley
?
Background
• DataSplash is a data visualization tool that
displays database data in graphical form
– Each row in a database table gets translated
into one graphical object on a canvas
DataSplash
Database
1.5232
2.8238
3.9221
Table
VIQING
Chris Olston, UC Berkeley
One row
Canvas
Example DataSplash
Visualization
• This visualization shows which political
party each state has favored since 1952
Each state is
one database
row
Red: Democrat
Blue: Republican
• A DataSplash canvas can be infinitely
panned and zoomed
VIQING
Chris Olston, UC Berkeley
Portals: Nested Visualizations
• Portals are sub-windows in one canvas that
show another canvas
Bush ‘88
Clinton ‘92
Dukakis ‘88
Bush ‘92
A Portal
This portal contains a
canvas of presidential
candidates ordered by
year (X axis), with the
winner on top (Y axis)
• Portals can be independently panned and
zoomed
VIQING
Chris Olston, UC Berkeley
Outline
• Introduction
– Related Work
– Background
•
•
•
•
Visual query results
Specifying visual queries
How VIQING generalizes other work
Status and future work
VIQING
Chris Olston, UC Berkeley
?
Visual Selection
• A visual selection displays only rows that
pass a selection filter
– Which states voted Democratic in 1992?
Note that all red
(traditionally Democratic)
states voted Democratic in
1992
VIQING
Chris Olston, UC Berkeley
?
Visual Join
• A visual join ( ) combines information
from two or more database tables via portals
Each presidential
candidate has a portal
containing the states
that voted for him
One join
portal for
every row in
the candidates
table
Presidential Candidates
VIQING
Chris Olston, UC Berkeley
States
Outline
• Introduction
– Related Work
– Background
•
•
•
•
Visual query results
Specifying visual queries
How VIQING generalizes other work
Status and future work
VIQING
Chris Olston, UC Berkeley
?
User Interface: Performing a
Visual Selection
• Select graphical rows by rubber-banding
• The result:
A portal that
contains only
the selected
rows
– The canvas inside the portal has only 6 rows
– Selection portals can be used for visual joins ...
VIQING
Chris Olston, UC Berkeley
?
Performing a Visual Join
• Drag . . . . . . . . and Drop
Join 1960’s
presidential
candidates with
political parties
VIQING
Chris Olston, UC Berkeley
?
The Result: A Three-Level
Visual Join
• Now candidates
are joined with
political parties
– We know which
candidates belong
to which parties
– Can see trends for
each party over
time
VIQING
Chris Olston, UC Berkeley
Parties
Candidates
States
?
Visual Reordering
• Visual queries
have an ordering
• Visual reordering
can be performed
after the join
– To reorder: drop a
portal onto a row
of its child canvas
VIQING
Chris Olston, UC Berkeley
Parties
Candidates
States
?
Result of Visual Reordering
• Now, parties join
with states, which
join with
candidates
Georgia voted with
the other Democrat
states in ‘60, but
against them in ‘64
– We can see the
voting history of
each state, by
traditional party
VIQING
Chris Olston, UC Berkeley
Parties
States
Candidates
Benefits of VIQING Queries
• Easier to use than SQL
– Can incrementally build and refine queries
– Query manipulations on custom graphical
representation of data, which is easier to
understand than text
– Don’t need to know SQL syntax -- just drag and
drop (direct-manipulation)
VIQING
Chris Olston, UC Berkeley
Join Predicates
• We have not discussed how VIQING knows
what join predicates to use
• In most cases, join predicates are equality
– eg, candidate.party_name = party.party_name
– These can be inferred from foreign key
relationships defined at schema creation time
• Alternatively, could specify more general
join predicates with a tool like MS Access
VIQING
Chris Olston, UC Berkeley
Removing Intermediate Tables
• Often, 2 tables join via an
intermediate table
– eg, Candidates
Vote records
92-TX-R
States
• However, we don’t want to see
the intermediate table
– we want
Candidates
States
• To do this, visually remove intermediate
– Drag intermediate portal away from the canvas
VIQING
Chris Olston, UC Berkeley
Outline
• Introduction
– Related Work
– Background
•
•
•
•
Visual query results
Specifying visual queries
How VIQING generalizes other work
Status and future work
VIQING
Chris Olston, UC Berkeley
How VIQING Generalizes Other
Work
• VIQING generalizes nested report writers
– Each level of nesting is a set of join portals
– Drill-down performed by entering a join portal
• VIQING generalizes master/detail forms
– Master-detail relationship is a join
– Data entry support could be added to
DataSplash
VIQING
Chris Olston, UC Berkeley
Generalizing “Small Multiple”
Graphs
• VIQING can create “small multiple” graphs
– Several views of a graph, indexed by a variable
Z=5
Z = 10
– This is a visual join between a canvas which
contains several values for the index variable
and the graph canvas
VIQING
Chris Olston, UC Berkeley
Status and Future Work
• Implemented as an extension to DataSplash
• Future work:
– Support for more SQL query expressibility
• aggregates, subqueries, etc.
– An automatic way to expose meta-data
• Which portals correspond to which tables?
– Improved support for large data sets
• This is a DataSplash issue, orthogonal to VIQING
VIQING
Chris Olston, UC Berkeley
Summary
• VIQING combines querying with
visualization by using portals
– Construct basic SQL queries by direct
manipulation of pictorial data
?
• Visual select, join, reorder, remove intermediate
– Create nested reports, master/detail forms
– Generate “small multiple” graphs
VIQING
Chris Olston, UC Berkeley
For more info...
• Paper in Proc. Visual Languages 1998
– Or my web page:
http://datasplash.cs.berkeley.edu/cao
• Email me: [email protected]
VIQING
Chris Olston, UC Berkeley