The paper was presented at the Social Networking Workshop, COMSNETS 2017

1. Mithileysh Sathiyanarayanan and Donato Pirozzi
City, University of London, UK
Mithileysh.Sathiyanarayanan@city.ac.uk
Social Networking Workshop, COMSNETS 2017
Bangalore, India
Social Network Visualization: Does Partial
Edges Affect User Comprehension?

2. Purpose of Social Network
Visualization
 Identify actors in a network
 Understand various relationships in a network
 Spot aggregate information in a network

3. Motivation
 Since these days people communicate everyday via
social networks, there is a big need for clutter-free
visualisation tools for the data analysts.
 However, for small-scale networks, the cluttered form of
set visualisation with graph motivates people to find
alternative ways to represent social networks in an
effective way.

4. What is the challenge?
In this research, the challenge is to understand whether
partial edges in graph help in better understanding of
visualizing a social network using an Euler diagram
augmented with Network diagram to identify social
relationships effectively.

5. Why Euler Diagram?
For its set relationship properties
1. Set Inclusion
2. Set Exclusion
3. Set Overlap

6. Why Network Diagram?
For its graph relationship properties
1. Graph node relationships
2. Graph edge relationships
3. Graph node-edge relationships

7. Why Euler Diagram + Network
Diagram?
We can augment both to get set and graph
relationships.

8. Addressing the Problem
 In a small-world network, Euler diagrams are augmented
with graphs that can be realized in social network
analysis, and we addressed the visual clutter in this
work.
 One interesting way of reducing visual clutter is to use
partial edges when there are visual elements (edge and
curve) crossings.
 We designed various combinations of diagrams and
conducted empirical evaluation.

9. Aim of the Study
The main aim of the study was to understand
“whether partial edges affect user comprehension in
social network visualisation?”

10. Design Specifics
 2 [diagram types] × 3 [question types] × 3 [diagram variations]
gave lead to 18 diagrams for each participant.
 2 Diagram Types
 partial and complete edges.
 diagrams were drawn using the software Inkscape.
 three different sizes of curves/rectangles were considered:
small, medium and large.
 3 Diagram Variations
 0 deg, 90 deg and 270 deg

11. Design Specifics (contd.)
 3 Question Types
1) How many people interested in COMPUTERS are
directly connected to someone who is not interested in
COMPUTERS?
2) How many direct connections are there between people
interested in BUSINESS and people not interested in
BUSINESS?
3) How many subjects include someone who is directly
connected to someone interested in SCIENCE?

12. Design Guidelines and
Characteristics
 To be consistent with other researchers, we
considered the following:
 For Euler diagrams, 10 guidelines were
considered.
 For Network diagrams, 3 guidelines were
considered.
 For the combination, 10 layout characteristics were
considered.

13. Example of the Diagrams

14. Study Approach
 20 participants were randomly selected for the study.
 Fourteen of the participants were male, six female, and
their age varied from 24 to 39 years with 28 years as the
mean value for their age.
 Next, all participants were asked questions based on the
visualisations.
 Diagrams drawn were randomised for the participants.
 User Study Lab
 Research Vehicle Tool

15. Results
 Error Analysis:
 Of the 360 observations, there were a total of 46 errors
(error rate of 12.77%). For partial edges, there were 17
errors (error rate of 9.44%) and for non-partial, 29 errors
(error rate of 16.11%).
 From the chi-square test, the p-value is 0.006, which
means there is a borderline significant difference between
partial edges and non-partial edges in terms of accuracy
(because p < 0.005).

16. Results (contd.)
 Time analysis:
 Of the 360 observations, the average mean time of each
question was 43.05 seconds. For partial edges, the average
mean time of each question was 41.86 seconds and for
non-partial, 44.24 seconds.
 So, from the ANOVA test, the p-value is 0.052, which
means there is a borderline significant difference between
partial edges and non-partial edges in terms of response
time (because p < 0.005).

17. Results (contd.)
 Preference Analysis:
 Of 20 participants, 16 participants (80%) preferred
partial edges based on the graph reading in Euler
diagrams and the layout.
 Participants feel partial edges are more aesthetically
pleasing over the traditional non-partial edges.

18. Theoretical Support
 One of the Gestalt principles on the diagrams’ is the
closure principle.
 This principle asserts that “incomplete or partial
obscured elements tend to be completed by the mind”
 So, the conclusion is “partial edges don’t affect user
comprehension in social network visualisation”

19. Future Works
 Other alternative ways of social network representation
based on Euler diagrams and Network diagrams are
considered for addressing the scalability issue.
 The existing Euler diagrams with complete graphs (full
edges) tool will be extended to partial edges and help
in exploring Euler diagrams and graph choices in a
dynamic and interactive platform.

20. Future Works (contd.)
 We plan to include a more complex scenario or a real
small-world example with complicated tasks focussing
on clustered network with complex set relationships
 In the next study, we will use crowdsourcing for
evaluation of the visualisations since the results
obtained with 20 participants of this study must be
verified for larger groups of participants.

21. Thank You
and
Any Questions?