Query by Attention:
Visually Searchable Information Maps
Mark A. Foltz
Randall Davis
MIT Artificial Intelligence Lab
July 25, 2001

Automobiles

Slide 3

Outline
Query by Attention
Review of Spatial Selection and Visual Search
Mapping Info Dimensions onto Perceptual Dimensions
Examples
Related Work, Conclusion

Perceptual Filtering

Information Filtering

Why Query by Attention?
User controls attention, instead of manipulating an interface
Rapid adjustment of query parameters
Immediate feedback
Can lead to more complete exploration of info space

Good Perceptual Dimensions
Preattentive map
Attended items are immediately available
Symmetric
Can filter for each kind of item equally well
Many values
Can code information precisely

Good Perceptual Dimensions
Spatial selection
Attention directed to contiguous regions
Color
Shape
Depth
Motion (not always symmetric)
Size

Spatial Selection

Spatial Selection

Spatial Selection

Spatial Selection

Spatial Selection

Color

Shape
Many factors influence the efficiency of search
line terminations, closure, holes, intersections
A conservative list
A less conservative list

Efficient Conjunctive Searches
Color and Shape
Egeth et al. 1984
Motion and Shape
McLeod et al. 1988, McLeod et al. 1991
Motion and Orientation
Driver and McLeod 1992
Motion and Depth
Nakayama and Silverman, 1986

Good Perceptual Dimensions
Spatial selection
Precise
Can query by range
Hierarchical
Color
Nine values
Used discretely here
Shape
Structure of this dimension poorly understood

Mapping Information to Perception

Mapping Information to Perception

Mapping Information to Perception

Mapping Information to Perception

Mapping Information to Perception

Mapping Information to Perception

Mapping Information to Perception

Slide 26

Mutual Funds

Mutual Funds

Related Work
Interactive information filtering
Ahlberg and Shneiderman, CHI 1994
Spotfire www.spotfire.com
Studies of user attention
Pirolli et al., CHI 2001
Presentation design systems
VQE (Derthick et al. UIST 1997)

Future Work
Handling large numbers of items
IMDB: 12,511 films 1980-2001
Let user filter first
Handling many information attributes
Multiple views, dimension reduction
Making it easy to create these kinds of maps
Assisting or automating info design

Conclusion
People perceptually filter.
Take advantage to filter abstract information.
A method to map information dimensions to perceptual dimensions.