Visualizing large connected graphs using level-of-detail and multi-resolution techniques Dr. David Breen Michael Kozak, Laurence Chapman Jeffery Marker, Mark Zuber

1
Visualizing large connected graphs using
level-of-detail and multi-resolution techniques
Dr. David BreenMichael Kozak, Laurence
ChapmanJeffery Marker, Mark Zuber
ACIN Phase 5
2
Outline

• Purpose
• Techniques
• Abstraction based techniques
• Distortion based techniques
• Graph decimation based techniques
• Level of detail reduction using FADE algorithm
• Reduction based techniques
• Zoom based techniques
• Software Packages tested

3
Purpose

• Part of COMPOSER is the ability to display
  thousands of nodes communicating simultaneously.
• Seemingly little effort has been put into
  examining how such a large number of nodes can be
  displayed.
• For the user, being able to quickly view and
  analyze the scene is critical to the success of a
  mission and a key part of COMPOSER.
• Thousands of nodes being displayed at once
  clutters the screen, rendering it impossible to
  analyze the situation quickly.
• Therefore, a detailed study into level of detail
  and multi resolution techniques on large graphs
  must be conducted in order to analyze possible
  solutions to this daunting problem in order to
  provide the best options for the user to choose
  between

4
Purpose (cont.)

• In short, how does one turn this

into something useful?
5
Examples of large graphs
6
Techniques

• Abstraction based techniques
• Distortion based techniques
• Graph decimation based techniques
• Level of Detail reduction using FADE algorithm
• Reduction based techniques
• Zoom based techniques

7
Abstraction based

• Sacrifice accuracy in exchange for readability
• Focuses on taking data from the scene and
  transforming it into trees, charts, tables, etc.
• Using this method destroys the graph, and with it
  the locations of nodes
• Not suited for situations when the user has to
  keep track of node locations

8
Distortion based

• Distort the scene to emphasize important details
• Both displays the entire graph and focus on a
  specific area
• Much like the zoom-based techniques, the nodes
  furthest away from the area of interest are
  usually hard to read.
• However, the variability in both the zoom and
  distortion levels can allow for flexibility in
  how to view the scene

9
Graph decimation based techniques

• Works on the basis that not all individual nodes
  are important, but an overall picture is
  necessary
• Takes clumped nodes and converts them into a
  single point
• A graph of thousands of nodes could be reduced to
  a few hundred or even a few dozen
• Cannot be used to determine if any two nodes can
  communicate with each other
• Allows the user to get a much better idea of how
  strong the network is overall, along with the
  structure of the node placement

10
LOD reduction using FADE algorithm
Base graph 21780 vertices 64266 edges
4 levels down 1542 vertices 7170 edges
7 levels down 39 vertices 150 edges
11
Reduction based

• Removes portions of the graph that are not
  necessary at the moment
• For example, if the user clicks on a node, only
  lines connecting that node and nodes it can
  communicate with be drawn
• Substantially reduces clutter in a scene,
  allowing for much better readability
• In a situation where the user needs to switch
  between nodes rapidly to examine the entire scene
  this method isn't recommended

12
Zoom based

• Focus solely on zooming into and out of the scene
• May zoom into the entire scene, or have a
  separate window that acts as a magnifying glass
  for a selected portion of the graph
• It does not allow the user to get an idea of the
  entire graph
• Even with a separate view port, it does not clear
  the clutter of having many hundreds or thousands
  of lines on screen

13
Software packages tested

• aiSee - Designed around fisheye and Cartesian
  coordinate views to allow for a large connected
  graph to be distorted in real-time
• Guess - Focuses on graph zooming and
  interactivity between the user and the nodes
• Large Graph Layout (LGL) Color coded graphing
  system based on zooming techniques
• Pajek - Offers an extremely robust visualization
  package encompassing many techniques including
  distortion, abstraction, and zooming.

14
Software packages tested (cont.)

• Prefuse - has a number of different graph views
  including a radial graph explorer that abstracts
  the scene to allow for dynamic movement of nodes
  and a hyperbolic tree view takes advantage of
  fisheye distortion to focus on any selected node
• VGL Uses node grouping along with abstraction
  and zooming to display and cluster graphs
• Walrus - Walrus uses abstraction based
  techniques, specifically hyperbolic geometry, to
  display graphs under a fisheye-like distortion
• Wilma - Through the use of cluster forces the
  user can force properties onto the graph to
  dynamically change the scene

15
aiSee (Regular graph)
16
aiSee (Cartesian view)
17
aiSee (Fisheye view)
18
Guess (Regular view)
19
Guess (Zoomed view)
The information window displays what 2 nodes are
connected
20
Guess (Abstracted view)
Guess has a number of ways to abstract graphs to
better examine a scene. Here, a radial view is
used to view connections branching out from a
specific node
21
LGL (Regular graph)
View Port
Node directory
22
LGL (Zoomed View)
3 levels of zoom
6 levels of zoom
23
LGL (Colorized view)
The vertices, edges, and background can be given
separate colors
24
Pajek (Regular view)
25
Pajek (Abstracted view)
These are just a few of the many possible
arrangements
26
Prefuse (Regular view)
27
Prefuse (Abstracted view)
Any node can become the center of the tree. The
graph is dynamically re-arranged to conform to
that node
28
Prefuse (Fisheye view)
As the graph is rotated, edges expand/shrink to
conform to the circular boundary
29
VGL (Regular view)
Double clicking on any node will bring up its
properties
30
VGL (Node clustering)
Any combination of nodes can be selected and made
into a group. All connections going to any of the
nodes in the group will go to the new, larger
group node.
31
VGL (Abstracted view)
Any node can be the root of the tree, and the
tree can be drawn down, left, right, or up
32
Walrus (Regular view)
The scene in Walrus can be dynamically rotated
and zoomed to provide the best angles for the user
33
Walrus (Reduced view)
This is the same graph as before, only the tree
has been trimmed to within 9 hops from the root
34
Wilma (Regular view)
Force manipulation window
View port
35
Wilma (Expanded view)
Adjusts the lighting on the nodes
36
Wilma (Distorted view)
Activating the FADE force will force the graph to
organize itself more like a mesh