Causality Visualization Using Animated Growing Polygons

1
Causality Visualization Using Animated Growing
Polygons

• Niklas Elmqvist (elm_at_cs.chalmers.se)
• Philippas Tsigas (tsigas_at_cs.chalmers.se)

IEEE 2003 Symposium on Information
VisualizationOctober 19th-21st, Seattle,
Washington, USA.
2
Outline

• Introduction and Motivating Example
• Related Work
• The Growing Polygons Technique
• User Study Results
• Conclusions Future Work

Roadmap
3
Introduction
Since we believe that we know a thing only when
we can say why it is as it iswhich in fact means
grasping its primary causes (aitia)plainly we
must try to achieve this ... so that we may
know what their principles are and may refer to
these principles in order to explain everything
into which we inquire. -- Aristotle, Physics
II.3.

• The concepts of cause and effect are pervasive in
  human thinking
• Causality is a very important reasoning tool in
  both science as well as everyday life
• Causal relations can be very complex
• This talk describes effective ways of visualizing
  causality

4
Example Citations

• Lets study the chain of citations in a
  collection of scientific papers
• A citation can be seen as an influence
• Citation graphs can be very large
• Studying these chains can give the following
  information
• How are authors are influenced by other authors?
• How are ideas propagated in a scientific
  community?

5
Example Citations (2)
time
1999
2000
2001
2002
2003
6
Causality Visualization

• Formally, we are looking to visualize systems of
  causal relations
• Def The causal relation ? is a relation that
  connects two elements (events) x and y as x ? y
  iff x influences y.
• Sets of events are called processes P1,..., PN
• Internal events are sequential and causally
  related
• External events interconnect processes through
  messages
• Effective visualization is a difficult problem
• Traditional visualization Hasse diagrams

7
Applications

• General information flow problems
• Rumor spreading
• Citation networks
• Software visualization
• Learning, designing, or debugging distributed
  programs and algorithms

8
Related Work Hasse Diagrams

• Distributed system with n20 processes and 60
  system events
• Difficult to comprehend
• Intersecting and coinciding message arrows
• Fine granularity
• The user must manually maintain the context of
  the relations
• Users may have to backtrace every single message
• Vital information is scattered

9
Related Work Growing Squares

• Our earlier attempt at improving causality
  visualization
• Processes represented by animated 2D squares
• Presented at SoftVis 2003
• More efficient than Hasse diagrams but
• Similar colors reduce scalability
• Influences are mixed up
• No absolute timing information

10
Growing Polygons

• Refinement of Growing Squares
• Idea Represent each process by an n-sided
  polygon (process polygon)
• Assign each process a unique color
• Assign each process a unique triangular sector in
  the polygons

11
Growing Polygons (2)

• Process polygons are laid out on a large n-sided
  layout polygon
• Each polygon grows as time progresses
• Animated timeline
• Messages are shown as arrows travelling from one
  process to another at specific points in time
• Messages carry influences (see next slide)

Simplified GP diagram
12
Growing Polygons Influences

• Messages carry influences (causal relations)
• Source color is transferred to the destination
• Causal relations are also transitive
• Transitive colors are also carried across
• Both color and orientation used for separating
  processes

13
Growing Polygons Example (1)
14
Growing Polygons Example (2)
15
Growing Polygons Example (3)
16
Growing Polygons Example (4)
17
Growing Polygons Example (5)
18
Hasse vs Growing Polygons
19
User Study

• A formal user study comparing Hasse diagrams to
  Growing Polygons was performed
• Two-way repeated-measures ANOVA
• Independent variables (both within-subjects)
• Visualization type Hasse or GP
• Data density sparse and dense
• 4 different data sets 1 of each data density for
  each visualization type
• 20 subjects participated in the test
• All subjects knowledgeable in distributed systems

20
User Study Tasks

• Each data set required the user to solve 4 common
  questions related to causal relations
• Find the process with longest duration
• Find the process that has had the most influence
  on the system
• Find the process that has been influenced the
  most
• Is process x causally related to process y?
• Times were measured for these tasks
• Users were also asked for their subjective
  opinion of the visualization (rating and ranking)

21
Results

• Performance measurement
• Users were more efficient using Growing Polygons
  than Hasse diagrams
• Hasse 434 (s.d. 379) seconds
• GP 252 (s.d. 175) seconds
• This is a significant difference for both sparse
  and dense densities

22
Results (2)

• Correctness
• Users are more correct when solving problems
  using Growing Polygons than Hasse diagrams
• Hasse 4.4 (s.d. 1.1) correct
• GP 5.6 (s.d. 0.7) correct
• This was a significant difference for both sparse
  and dense densities

23
Results (3)

• Subjective ratings
• Very positive user feedback
• Users consistently rated GP over Hasse diagrams
  in all respects (ease-of-use, enjoyability,
  efficiency)
• These readings were all statistically significant
• The majority of users also rated GP over Hasse

24
Conclusions Future Work

• Visualization of causal relations is crucial for
  understanding complex systems
• Traditional visualization techniques (Hasse
  diagrams) fall short
• Growing Polygon is a novel idea of visualizing
  causality focused on the information flow
• Our visualization technique is
• Significantly more efficient to use than Hasse
  diagrams
• Significantly more appealing to users than Hasse
  diagrams
• In the future we want to explore scalability
  concerns in systems spanning long time periods
  and involving many processes

25
Questions?

• Contact information
• Address
• Niklas Elmqvist and Philippas Tsigas
• Department of Computing Science
• Chalmers University of Technology
• SE-412 96 Göteborg, Sweden
• Email
• elmtsigas_at_cs.chalmers.se
• Project website
• http//www.cs.chalmers.se/elm/projects/causalviz