Efficient%20User%20Interest%20Estimation%20in%20Fisheye%20Views - PowerPoint PPT Presentation

Title: Efficient%20User%20Interest%20Estimation%20in%20Fisheye%20Views


1
Efficient User Interest Estimation in Fisheye
Views

• Jeffrey Heer and Stuart K. Card
• 1 Palo Alto Research Center, Inc.
• 2 University of California, Berkeley

1,2
1
2
Roadmap

• Motivation Background
• Implementation
• Evaluation
• Conclusion

3
Fisheye Views
4
Degree of Interest (DOI)

• Models users spontaneous interest across the
  tree
• This model can then be used to inform presentation

0
0
-1
0
-1
0
0
-1
-2
0
-1
-2
0
0
-1
-3
-3
-3
-3
-1
-2
-2
Computed Degree of Interest
Cull low Degree of Interest
5
User Modeling in Fisheye Views
Degree of Interest ? Intrinsic Importance
Distance from Point of Interest

3
0
4
2
1
1
5
5
1
3
2
2
2
2
0
3
6
6
6
6
2
4
4
3
3
3
3
3
3
3
Intrinsic Importance
Distance from Point of Interest
6
DEMO
7
the need for speed

• Visualization should respond fluidly to user
  actions
• But for each interaction, may have to
• Recompute DOI
• Recompute Layout
• Hard time limit 100ms (Card, Moran, Newell)
• Goal
• Limit all computations to the number of displayed
  nodes.

8
Naïve Interest Computation

0

-1
0

-2
0
-1
-2

0
-3
-3
-3
-3
-1
-2
-2
Requires visiting the entire tree!
9
Least Common Ancestor Pruning
Limit computation to the subtree rooted at least
common ancestor.
0
-1
0
0
-2
0
-1
-1
-2
0
-3
0
-2
-3
-3
-3
-1
-2
-2
-2
0
-1
-3
However, no savings if new focus is here
Furthermore, this method exploits a specific DOI
distribution ? not necessarily generalizable
10
Solution Disinterest Thresholding

• Saturate DOI function at a disinterest threshold
• Compute DOI only for visible nodes
• Use thresholding to supply defaults for the others

0
0
-1
0
-1
0
0
-1
0
-1
-2
-2
0
-2
-2
-2
-2
0
-1
-1
-2
-2
Cull low Degree of Interest
Computed DOI minDOI -1
11
Disinterest Thresholding
0
-1
0
0
-1
-2
-2
-2
-2
-2
-2
0
-1
-2
-2
12
Node Attribute Registry

• Backing array data structure table of node
  attributes.
• Tag visible nodes with table index. When
  attributes are needed (e.g. node.getX()), the
  table is consulted.
• If the node is in the table, the attribute is
  simply returned.
• Else, the suitable default is supplied
• DOI minimum DOI, Position position of first
  visible ancestor

index dirty DOI x y size color etc
0 1 0 213 12 5 ..
1 1 -1 134 58 4 ..
13
Evaluation
Setup Time walks through algorithmically
generated DOITrees with increasing tree
depths. Test System PIII 1GHz, 256MB RAM 16 MB
Video RAM DOI Threshold -2
Naïve and LCA grow linearly with the number of
nodes. Disinterest thresholding grows linearly
with number of visible nodes, which in this case
grows logarithmically with total number of nodes.
14
Limitations

• Doesnt improve cases where there are a large
  number (10,000) visible nodes.
• Smooth interaction also dependent on the use of
  efficient layout algorithms.
• Only approximates DOI distribution, which may be
  problematic if applications wish to use DOI for
  more than visualization.

15
Thanks!! Questions?

• Jeffrey Heer jheer_at_parc.com
• Stuart K. Card card_at_parc.com

16
Motivation

• The real design problem is not increased access
  to information, but greater efficiency in finding
  useful information.
• Increasing the rate at which people can find and
  use relevant information improves human
  intelligence.

17
Information Visualization

• Leverage highly-developed human visual system to
  achieve rapid understanding of abstract
  information.

1.2 b/s (Reading) 2.3 b/s (Pictures)
18
Node Attribute Registry

• DOI function only sets DOI for nodes above the
  disinterest threshold.
• Nodes are transparently added to registry when
  DOI is set.
• If node is already there, then dirty bit is set.
• Registry is resized as necessary.
• After DOI computation, non-dirty nodes are
  removed from registry, dirty bits are cleared.
• Result DOI computation time proportional to the
  number of nodes displayed!

19
Possible Questions

• What about other DOI distributions?
• examples?