Transcript Interactive Linking as Query Specification

SIMS 247 Lecture 6
Linked Interaction as Query
Specification
February 5, 1998
Marti Hearst
SIMS 247
Today
• Discuss Assignment 1
• Introduce Assignment 2
• Iterative Visualization as
Query Specification
Marti Hearst
SIMS 247
Dynamic Queries
• Instead of a formal database language
• Explore a dataset interactively
• Use graphical devices to interactively
update a visualization
– Examples
• Ahlberg & Shneiderman 93 Filmfinder, etc.
• Roth et al. 96 VISAGE
• Woodruff et al. DataSplash
Marti Hearst
SIMS 247
Filtering / Restricting
• Select a subset of data to be filtered out
(or retained)
• How does this differ from highlighting?
– Highlighting shows relationships between one
subset of points and the full set
– Filtering removes a subset of points from
consideration
Marti Hearst
SIMS 247
Interaction as Query Specification
Interactive manipulation of a
visualization can be seen as a kind of
query specification
– Filters as ANDs
e.g., Show only records with employees who
make more than $30k. Now show the cities
they live in. Many cities omitted.
– Alternative views as ORs
Marti Hearst
SIMS 247
Tight Coupling
• Tight coupling: like brushing and linking, with
an emphasis on use of filters
• Link together two or more query components
to preserve display invariants
– Example from standard GUI:
• Grey out inapplicable menu choices
• Helps prevent user from illegal actions
• For information exploration
– Output of one query -> input to another
Marti Hearst
SIMS 247
Tightly-Coupled Visualization for
Query Refinement
• Progressive refinement
– users can see if they are going to get too
many results
– adjust query if so
• Easier exploration
– users quickly undo a constraint imposed by
a filter and try a different path
Marti Hearst
SIMS 247
VISAGE (Roth et al. 96)
• Tightly coupled database queries
– drag & drop data subsets
– enforce constraints automatically
• Multiple, complementary ways to present and
view information
• Data-centric approach
– indicate operations on the data itself
– as opposed to selecting from disassociated menus
• no need to gray out inappropriate menu choices
– as opposed to “file-centric” or “document-centric”
Marti Hearst
SIMS 247
Dimensions for interaction (Roth et al. 96)
• Set selection
• which objects to work on
• how to indicate attributes and ranges
• Granularity/composition of actions
• how to indicate what operation to perform
• how to combine several operations
• granularity of operations
– automatic breadmaker vs. bowls, rolling pin, etc.
• Continuity of action
• adjusting light dimmer vs. picking tv channel
• Consistency with domain
• how do people usually interact with this kind of
information?
Marti Hearst
SIMS 247
Features of VISAGE (Roth et al. 96)
• Works with an underlying database
• User can manipulate individual data records
or metadata for tables
• Context-sensitive menus allow for access to
finer-granularity information (“drill down”).
Marti Hearst
SIMS 247
VISAGE display (Roth et al. 96)
Marti Hearst
SIMS 247
VISAGE display (Roth et al. 96)
Marti Hearst
SIMS 247
VISAGE display (Roth et al. 96)
Marti Hearst
SIMS 247
VISAGE features
• Data-centric approach
– make choices on the data
– as opposed to selecting from disassociated menus
– as opposed to “file-centric” or “document-centric”
• other systems have had this focus
– e.g., smalltalk object programming system
Marti Hearst
SIMS 247
Other VISAGE features
• Menu of applicable attributes is
attached to the frame of the bar chart
• Icons used are a convention for the
domain
• Copying a visual representation always
refers to the underlying data point
– as opposed to drawing programs
– the visual appearance of the paste
operation is based on the kind of graph the
data is dropped into
Marti Hearst
SIMS 247