Network Visualization by
Semantic Substrates
Aleks Aris
Ben Shneiderman
Outline
• Graph Drawing Aesthetics
• Node Placement Methods
• Challenges of Network Visualization
• Semantic Substrates
• Demo
• Questions & Answers
Graph Drawing Aesthetics
•
•
•
•
•
•
•
•
•
•
•
Source: [9] Davidson & Harel
Minimize edge crossings
Draw links as straight as possible
Maximize minimum angle
Maximize symmetry
Minimize longest link
Minimize drawing area
Centralize high-degree nodes
Distribute nodes evenly
Maximize convexity (of polygons)
Keep multi-link paths as straight as
possible
…
Node Placement Methods
• Node-link diagrams
– Force-directed
– Geographical maps
– Circular layouts
• One or multiple concentric
– Temporal layouts
– Clustering
– Layouts based on node attributes (later)
• Matrix-based representations
Force-directed Layout
• Also known
•
as: Spring
Spreads nodes
– Minimizes
chance of
node
occlusion
Source: www.visualthesaurus.com
Geographical Map
SeeNet, Becker et al.
• Familiar
location
of
nodes
Source: www.visualcomplexity.com (100)
Circular Layouts (1 circle)
Schemaball, Martin Krzywinski
• Ex: Schemaball
– Database schema
– Tables connected
via foreign keys
Source: http://mkweb.bcgsc.ca/schemaball/?home
Circular Layouts (concentric)
Radial Tree Viewer, Nihar Sheth
Source: www.visualcomplexity.com (26)
Circular (concentric) & Temporal
Hudson Bay Food Web
Source: http://www.itk.ca/environment/tek-ecological-foodweb-image.php
Temporal Layout
Source: [16] Garfield, “Historiographic mapping of knowledge…”
Clustering
Vizster, Heer et al.
Source: www.visualcomplexity.com (85)
Hierarchical Clustering
Source: [33] Schaffer, et al.
Matrix-based Layout
VisAdj, Ghoniem et al.
• Ex: VisAdj (Ghoniem, et al.)
– column: source vertex
– row: target vertex
Source: [17] Ghoniem et al., IEEE 0-7803-8779-1/04
Challenges of Network Visualization
• C1) Basic networks: nodes and links
• C2) Node labels
– e.g. article title, book author, animal name
• C3) Link labels
– e.g. Strength of connection, type of link
• C4) Directed networks
• C5) Node attributes
– Categorical (e.g. mammal/reptile/bird/fish/insect)
– Ordinal
(e.g. small/medium/large)
– Numerical (e.g. age/weight)
• C6) Link Attributes
– Categorical (e.g. car/train/boat/plane)
– Ordinal
(e.g. weak/normal/strong)
– Numerical (e.g. probability/length/time to traverse/strength)
C1) Basic Networks (nodes & links)
Power Law Graph, Linyuan Lu
• Power Law Graph
– 5000 nodes
– Uniformly distributed
Source: www.visualcomplexity.com (135)
C1) Basic Networks (continued)
Vizster, Heer et al.
• Social friendship network
– 3 degrees from Heer
– 47,471 people
– 432,430 relations
Source: www.visualcomplexity.com (97)
C2) Node Labels
Internet Industry Partnerships, Valdis Krebs
Source: www.visualcomplexity.com (168)
250 nodes
• Adding labels
– Nodes overlap with other nodes
– Nodes overlap with links
C3) Link Labels
Netscan, Marc Smith
• Challenges:
–
–
–
–
Source: www.visualcomplexity.com (127)
Length
Space
Belongingness
Distinction
from other
labels & other
types of labels
C4) Directed Networks
Yeast Protein Interaction
• Direction
– arrows
– labels
– Thickness
– color
SeeNet, Becker et al.
Source: www.visualcomplexity.com (127)
Source: [1] Becker et al.
C5 & C6) Node & Link Attributes
CIA World Factbook Visualization, Moritz Stefaner
Source: www.visualcomplexity.com (192)
• Value of node attribute indicated by node shape
• Value of link attribute indicated by a letter
Semantic Substrates
• Group nodes into regions
– according to one attribute
•
• Categorical, ordinal, or binned numerical
In each region:
– place nodes according to other attribute(s)
• Advantages
– Location conveys meaning, interpretable
– Instant perception of
•
• different types of nodes
• their relative number
• connections between different groups of nodes
Limitations
– Beyond 5 regions becomes challenging
– Constraint on nodes interferes with aesthetics
Demo
Thank you!
Questions & Answers
NVSS: Network Visualization by
Semantic Substrates
Comparison with FruchtermanReingold
Filtering Links by source-target
Filtering links by time attribute (1)
Filtering links by time attribute (2)
Overlapped Links
Three Regions
• Links from District
•
Courts
Indicates longevity
of cases (short to
long)
– District
– Circuit
– Supreme
Scalability
• 1280x1024
• 1,122 nodes
• 7,645 links
Using a third attribute in regions
• 13 circuits for
both Circuit
and District
Courts
• Horizontally
separated
• Reveals that
links remain
mostly within a
circuit
although there
are some
across (lateral
citations)