Graph Visualization: A Survey

Graph Visualization

Overview
 Survey is focused on the use of graph visualization techniques to

 Applications, Key Issues

problems in Information Visualization.

 Graph Layout

 Graphs - a natural way to represent relationships.

 Navigation and Interaction

 Graphs in Infoviz: Are there inherent relationships among the input

 Clustering

data elements?

 Systems

NO: Unstructed Data - Infoviz must be used to discover relationships.

 Current State of Graph Visualization

YES: Graphs can be used to represent the relationships - data

Reference

elements (nodes) and edges (relationships).

I.Herman, G. Melancon, M.S. Marshall, Graph Visualization and Navigation in Information Visualization: A Survey, IEEE Transactions on
Visualization and Computer Graphics, Volume 6(1), January, 2000.
1

ITCS 6010/8010: Information Visualization

Graph Visualization

 Survey is focused on structured data and the use of graphs applied

to Infoviz domain.

ITCS 6010/8010: Information Visualization

Graph Visualization

Applications
Key Issues in Graph Visualization
 File hierarchies, organization charts, biological taxonomies
 Web site maps, browsing history

 Size is the key issue in visualizing graphs.

 Biology: Evolutionary, phylogenetic trees, genetic maps, biochemi-

 Large graphs are constrained by display resources, and usability

cal pathways, protein function

 Computer Science: data structures, object oriented systems, ER

(interaction, for instance)

 Very little work on applying cognitive science or human factors.

diagrams, scene graphs (graphics)

ITCS 6010/8010: Information Visualization

Graph Visualization

ITCS 6010/8010: Information Visualization

Graph Visualization

Example Tree Layout

Graph Layout Algorithms

Reingold and Tilford Algorithm

Background: Graph Drawing

Given a set of nodes and a set of edges (relations), calculate the position
of the nodes and the curve to be drawn for each edge.

 Well studied subject (hundreds of publications, books!)

 Planarity: Graphs drawn on a plane with no edge crossing
 Layout types: grid layouts (nodes at integer coords), force-directed
models, simulated annealing based layouts.

 Aesthetic Issues: Nodes, edges evenly distributed, isomorphic substructures, straight line edges, etc.

Isomorphic subtrees are always laid out the same way

Distance between nodes is a parameter.

ITCS 6010/8010: Information Visualization

Graph Visualization

ITCS 6010/8010: Information Visualization

Graph Visualization

Layout Algorithm Issues

 Size: Graphs with thousands of nodes can make most layout algo-

Layout Algorithm Issues (contd)

rithms almost useless.

 NicheWorks, H3Viewer are among systems that can handle large

graphs.

 Predictability: Consecutive runs of an algorithm should not result in

radically different visual representations (preserving mental map)

 High Node Density: Traditional algorithms cannot scale up, making

interaction difficult.

 Time Complexity: Real-time (or near real-time) updates of the graph

is needed in highly interactive applications, where the graph can be
modified.

 Can also use 3D layouts or non-Euclidean geometry.

 Extremely large graphs - need methods to simplify (reduce) graph
size (clustering, grouping)

ITCS 6010/8010: Information Visualization

Graph Visualization

ITCS 6010/8010: Information Visualization

Graph Visualization

Traditional Layout
 Child nodes below common ancestor
 Reingold and Tilford algorithm - the best known - can produce topdown or lef to right layouts, as well as grid positioning.

 H-Tree: Representation for binary trees; uses axis-aligned representaton of edges.

 Radial Trees (Eades)

 Balloon Views - planarized Cone Trees
 Tree maps, Onion Graphs, Cushion Treemaps

ITCS 6010/8010: Information Visualization

Graph Visualization

ITCS 6010/8010: Information Visualization

10

Graph Visualization

Traditional Layout (contd)

 Spring Layouts: or force-directed
Models nodes and edges as physical bodies tied with springs.
Uses Hookes law to define forces between bodies
Can lead to well balanced layouts, some optimzed for minimizing
edge crossings and predictable layouts.
Disadvantage: Computational complexity is generally O(N 3),
most methods are highly unpredictable.

ITCS 6010/8010: Information Visualization

11

Graph Visualization

ITCS 6010/8010: Information Visualization

12

Graph Visualization

Spanning Trees

3D Layouts

 Most traditional layout algorithms work well on small graphs

 A practical solution is to layout the graph by computing the spanning
tree (tree layouts are typically of O(N ) complexity).
 Once the tree is laid out, the other edges can then be added to tree.
 Methods:

 3D layouts, in combination with interactive viewing can accommodate larger graphs, while compensating for occlusion

 Most force-directed and simulated annealing approaches can also

be extended to 3D.

 3D layouts also use visual cues such as transparency, depth queu-

Can use breadth-first or depth-first search, starting from a likely

root node.
If edges have weight, can use various optimization criteria (and
heuristics) in the layout algorithm

ITCS 6010/8010: Information Visualization

13

Graph Visualization

ing, etc.

 Classic Example: Cone Trees (a direct 3D algorithm) and their extensions (disk trees)

ITCS 6010/8010: Information Visualization

15

Graph Visualization

Cone Trees

3D Radial Layout

ITCS 6010/8010: Information Visualization

14

Graph Visualization

ITCS 6010/8010: Information Visualization

16

Graph Visualization

Other 3D Layout Examples

Hyperbolic Layouts

 SGI file system navigator (until version 5) (seen in Jurassic Park!)

 Perspective Wall - data representation as posters on large walls
 Viznet, Vitesse - data mapped onto a sphere

views

 Makes it possible to view and interact with large trees, and thus crucial for real-life Infoviz applications.

Difficulties with 3D layouts

 A mapping from hyperbolic to Euclidean space - two models, Klein

3D navigation in 2D displays and 2D input devices

and Poincare.

VR and immersive environments are not commonplace

ITCS 6010/8010: Information Visualization

 Provides a distorted view of a tree in 2D or 3D, similar to fish-eye

17

Graph Visualization

ITCS 6010/8010: Information Visualization

18

Graph Visualization

Hyperbolic Layouts(contd)
Hyperbolic Layouts(contd)
Hyperbolic Layouts(contd)
 Klein Model:
Klein model uses an open disc (sphere in 3d) as a subset hyperbolic plane contains points internal to the disc
Line segments of equal length in hpyerbolic plane get exponentially smaller as they approach the perimeter

 Poincare Model: Similar to the Klein model, but segments are arcs,
intersecting orthogonally at the perimeter of the disc.

ITCS 6010/8010: Information Visualization

19

Graph Visualization

ITCS 6010/8010: Information Visualization

20

Graph Visualization

Hyperbolic Layouts(contd)
Navigation and Interaction
 Crucial tools to dealing with large graphs
 Can help reveal graph structure
 Real-time interaction is an important goal.
The small wedges seen in Euclidean plane (resulting in unusable
layouts is very different in the hyperbolic plane (wedges are opened
up)

ITCS 6010/8010: Information Visualization

21

Graph Visualization

ITCS 6010/8010: Information Visualization

22

Graph Visualization

Space-Scale Diagrams
 Represent points in image as rays (with point, magnification)
 Choose paths to destination as a combination of zoom and pan ac-

Zoom and Pan

tions

 Standard graphics interaction operations, manipulated by affine

transformations, followed by a redraw.

 Geometric zoom vs. Semantic Zoom.

 Zooming in, can cause loss of context

ITCS 6010/8010: Information Visualization

23

Graph Visualization

ITCS 6010/8010: Information Visualization

24

Graph Visualization

Fisheye Distortion(contd)
Focus+Context Techniques
Techniques that allow users to focus on some detail of the visualization,
without losing context.

Fisheye Distortion
 Simple distortions use a radial function, for eg.,
h(x) = (d + 1)/(d + 1/x)

ITCS 6010/8010: Information Visualization

25

Graph Visualization

ITCS 6010/8010: Information Visualization

26

Graph Visualization

Fisheye Distortion(contd)
 Cartesian Fisheye: Distortion applied to X and Y independently.

Fisheye Distortion(contd)

 Approximation of segments with linear elements can introduce edge

crossings.

ITCS 6010/8010: Information Visualization

27

Graph Visualization

ITCS 6010/8010: Information Visualization

28

Graph Visualization

Focus+Context and Layout

Clustering

 Fisheye distortion is typically a processing step separate from the

 Advantageous to reduce the number of elements displayed, espe-

layout.

 Hyberbolic layout is the exception, as the technique is designed with

the goal of focus+context.

 Hyperbolic layouts - changing focus is equivalent to changing the

center of the Euclidean circle(sphere).

 Other Examples: mapping onto spheres and ellipsoids, followed by

perspective projection, Perspective Wall.

ITCS 6010/8010: Information Visualization

29

cially very large graphs/trees.

 Structure-based clustering: uses only structural information about

the graph

 Content-based clustering: node and edge attributes are used to determine clusters.

 Most clustering work is structure based.

 Clustering - used to accomplish filtering and search.

Graph Visualization

ITCS 6010/8010: Information Visualization

30

Graph Visualization

Cluster Layout
Clustering Approaches
 Can just layout the clusters by themselves - good high level view of
the graph structure

 Locate disjoint clusters (simpler to navigate)

 Choose clusters with the least number of edges between members
(Ratio-Cut technique)

 Use glyphs to represent clusters and treat them as supernodes

with abridgemens between clusters.

 Force-directed cluster layout: nodes that are related to each other

are attracted, while all nodes have a minimal repulsion force (O(N 3 )

 Hierarchical clustering

complexity).

ITCS 6010/8010: Information Visualization

31

Graph Visualization

ITCS 6010/8010: Information Visualization

32

Graph Visualization

Cluster Representation
Node Metrics for Clustering
 Structure or content based.
 Combination of structure and content based metrics are more powerful, eg. linear weighted combination.

 Representation of unselected nodes

Ghosting - deemphasizing nodes
Hiding
Grouping - grouping under a new super-node representation

ITCS 6010/8010: Information Visualization

33

Graph Visualization

Graph Drawing - Systems, Journals,

Conferences
 A very large community of researchers, users
 Large number of systems have been implemented
 Graph Drawing Symposia - dedicated to this domain
 Journal of Graph Algorithms and Applications
 Significant ties to CHIXX, UISTXX, InfoVizXX, IEEE TVCG, CG
Forum

ITCS 6010/8010: Information Visualization

35

Graph Visualization

ITCS 6010/8010: Information Visualization

34

Graph Visualization