Exploring social networks with
Matrix-based representations
Nathalie Henry
Co-supervisors:
Jean-Daniel Fekete & Peter Eades
Nathalie.Henry@lri.fr
Exploring social networks with matrices
Background
Peter Eades,
Univ. of Sydney
Sociologists
from EHESS,
Historians from the
French Archives,
Analysts from EDF
and France telecom
Jean-Daniel Fekete,
INRIA
Nathalie.Henry@lri.fr
Exploring social networks with matrices
2
Social Networks

What is it?


A set of actors connected by relations
a HOT topic




Online communities (FaceBook, Flickr)
Online collaboration (Wikipedia, Sourceforge)
Scientific collaboration
And more…



FaceBook Interactive Graph
Internet Traffic [Eick et al.]
The WEB
Diseases transmission networks
Terrorists networks
Nathalie.Henry@lri.fr
Exploring social networks with matrices
3
Analyzing social networks


Answering questions with statistics
Exploratory Data Analysis [Tukey,77]


Answering question you did not know you had
Looking at the raw data from different perspectives
Anscombe’s numbers
Nathalie.Henry@lri.fr
Statistics
Visual representation
Exploring social networks with matrices
4
Network Visualization

What social scientists start with:

Node-Link Diagram
 Overlapping nodes
 Edge-crossing
 Identify connections
100+ actors - InfoVis community
Nathalie.Henry@lri.fr
Exploring social networks with matrices
5
Network Visualization

What social scientists end with:

[Ke et al.,04]
Nathalie.Henry@lri.fr
Most important tasks
 Communities
 Central Actors
 Discover the
structure of the
network
Exploring social networks with matrices
6
Large Network Visualization

What social scientists start with:

Node-Link Diagram
 Overlapping nodes
 Edge-crossing
 Identify connections
4000+ actors
Nathalie.Henry@lri.fr
Exploring social networks with matrices
7
Large Network Visualization

What social scientists end with:
4000+ actors
Nathalie.Henry@lri.fr
Exploring social networks with matrices
8
What are the solutions?



A year of emails
Sampling, filtering
Clustering into meta-nodes
Alternative representations such
as adjacency matrices
A
A
C
B
Nathalie.Henry@lri.fr
B C D
1
1
B 10
0 1
0
C 10
10
D 10
0 10
A 0 1
D
0 1
0
Exploring social networks with matrices
9
My Approach

Participatory Design
Observation
Evaluation
Brainstorming
Prototyping

Outcomes



They use statistics, graph analytics
They use mainly node-link diagrams
But they use sometimes matrices
Nathalie.Henry@lri.fr
Exploring social networks with matrices
10
My Approach
Perception
Exploration
Information Visualization
Communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
11
Itinerary of my PhD
[Survey in progress]
1. Make
matrices usable
3. Augment
matrices
[Henry and Fekete, IHM06&Infovis06]
2. Combine
matrix+node-link
4. Merge
matrix+node-link
[Henry and Fekete, Interact07]
Interaction technique
[Elmqvist et al., CHI08]
[Henry et al., Infovis07]
Case study [Henry et al., IJHCI07]
Nathalie.Henry@lri.fr
Exploring social networks with matrices
12
Outline
Exploration
Perception
1. Make
matrices usable
3. Augment
matrices
2. Combine
matrix+node-link
4. Merge
matrix+node-link
Communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
13
1. How to make matrices usable?

J. Bertin, 1967
High-level structure?



Groups
Trends
Outliers
 Make them VISUAL
 REORDER their rows and columns
Nathalie.Henry@lri.fr
Exploring social networks with matrices
14
Reordering methods
[Survey to be published]

Lot’s of methods !



Table-based ordering methods
Graph linearization
Mine! (mixed approach)
A B C D E F G H I
Graph Linearization
Nathalie.Henry@lri.fr
J
Hierchical clustering
of microarray data
Exploring social networks with matrices
15
Mixed approach
[Henry and Fekete, IHM’06]
[Henry and Fekete, InfoVis’06]

Place actors with similar connection
patterns beside each other
ABCDEFGH
B
E
H
A
C
F
D
G
A
B
C
D
E
F
G
H
1
0
0
0
0
0
0 0 0 1
0 0 0 0
0
1
1
1
0
0
1 1
0 0
0 0
0 0
1 0
0 1
0 0 0
1 0 0
0 1 0
0 0 1
0 0 0
0 0 0
0
0
0
0
1
1
0 0 0 0
1 1 0 0
ABCDEFGH
A
B
C
D
E
F
G
H
1
2
2
0
3
3
2 3 3 1
3 2 2 4
0
1
1
1
2
2
1 1
0 2
2 0
2 2
1 3
3 1
2 2 2
1 3 3
3 1 3
3 3 1
0 2 4
2 0 4
3
2
2
4
1
1
4 4 0 5
1 1 5 0
 Add information to the adjacency matrix
Nathalie.Henry@lri.fr
Exploring social networks with matrices
16
Reordering methods
There are lots of methods…
… but the real question is:
 What is a GOOD ordering?
Nathalie.Henry@lri.fr
Exploring social networks with matrices
17
What is a good ordering?
… for performing the analysis of a social network
Most important
tasks

Communities:
B and C

Central Actors:
A
Manual ordering
Nathalie.Henry@lri.fr
Exploring social networks with matrices
18
What is a good automatic ordering?


Algorithms designed from formal measures
Can we characterize them according to the
visual features they produce?
Nathalie.Henry@lri.fr
Exploring social networks with matrices
19
Empirical study
[Henry and Fekete, Beliv’06]

How people
perceive groups?
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr
Exploring social networks with matrices
20
Is there A good ordering?

For a specific





visual feature?
type of data?
No :(
But it saves them time to start from somewhere!
Analysts need several orderings to find a consensus in the
data
 I focused on assisted reordering
and interactions to find consensus
Nathalie.Henry@lri.fr
Exploring social networks with matrices
21
Interacting to find a consensus
[Henry and Fekete, IHM’06]
[Henry and Fekete, InfoVis’06]
> interactive (fuzzy) clustering
> interactive ordering
Nathalie.Henry@lri.fr
Exploring social networks with matrices
22
Outline
Exploration
Perception
1. Make
matrices usable
3. Augment
matrices
2. Combine
matrix+node-link
4. Merge
matrix+node-link
Communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
23
2. Combining best of both worlds with MatrixExplorer
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr
Exploring social networks with matrices
24
MatrixExplorer
[Henry and Fekete, IHM’06]
[Henry and Fekete, InfoVis’06]

Coordinated views
Interaction to explore

Our hypothesis:




Matrix to explore
Node-link to communicate
Our observations:



Node-link for certain tasks, matrix for others
Cognitively demanding to switch back and forth
To find a consensus, they use both representations
different layouts/orderings
Nathalie.Henry@lri.fr
and
Exploring social networks with matrices
25
Exploring with matrix or node-link?

Social networks



Sparse (genealogical tree)  node-link
Large/Dense (a year of emails)  matrix
Why are users switching back and forth
when exploring large/dense networks?
Nathalie.Henry@lri.fr
Exploring social networks with matrices
26
Matrices weakness

The problem of path-related tasks[Ghoniem et al., 2005]


Always possible on matrices…
… but tedious !
?
?
?
A
A
B 0
C 0
Nathalie.Henry@lri.fr
1
A
0 1 0
B
0 0 1
C
0 0 0
A 0 1
C
B
B C D
D
D 0
1
Exploring social networks with matrices
27
Outline
Exploration
Perception
1. Make
matrices usable
3. Augment
matrices
2. Combine
matrix+node-link
4. Merge
matrix+node-link
Communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
28
3. Augmenting matrices with MatLink
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr
Exploring social networks with matrices
29
MatLink
[Henry and Fekete, Interact’07]
Best paper award



Adding static links
Adding interactive links
For larger matrices…
Nathalie.Henry@lri.fr
Exploring social networks with matrices
30
Mélange
[Elmqvist, Henry, Riche and Fekete, CHI’08]


Folding the space between 2 or more points of focus
Works well for matrices and social networks tasks
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr
Exploring social networks with matrices
31
Exploring with matrix or node-link?

Social networks

Sparse (genealogical tree)  node-link
Large/Dense (a year of emails)  matrix

Small-world networks:

A common category of social networks
 Globally sparse but locally dense
 node-link?  matrix?

Nathalie.Henry@lri.fr
Exploring social networks with matrices
32
Outline
Exploration
Perception
1. Make
matrices usable
3. Augment
matrices
2. Combine
matrix+node-link
4. Merge
matrix+node-link
Communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
33
Small-world

The best representation?

What is happening


Inside communities?
Between communities?
[Auber et al., InfoVis’04]
Nathalie.Henry@lri.fr
Exploring social networks with matrices
34
4. Merging matrix and node-link with NodeTrix
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr
Exploring social networks with matrices
35
Interactive pen tablet
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr
Exploring social networks with matrices
36
NodeTrix
[Henry et al., InfoVis’07]

Explore
QuickTime™ and a
decompressor
are needed to see this picture.
Nathalie.Henry@lri.fr

Communicate
QuickTime™ and a
decompressor
are needed to see this picture.
Exploring social networks with matrices
37
Actors in one or more communities

Where to place them?






Between communities?
In one or the other community?
In overlapping communities?
In a higher level community?
Why not duplicating them?
What are the effect of duplication
on user understanding?
Nathalie.Henry@lri.fr
Exploring social networks with matrices
38
Node Duplication


More accurate community view
Minimizing the misleading effect
by visualizing links between duplicates
Nathalie.Henry@lri.fr
Exploring social networks with matrices
39
What about evaluation?

Participatory design


Controlled experiment


For specific tasks (low-level) on specific datasets
with specific representations
Case Study


Mostly informal and during the whole process
More realistic settings
Longitudinal Study… coming
Nathalie.Henry@lri.fr
Exploring social networks with matrices
40
Case Study on 20 years of 4 HCI conferences
[Henry et al., IJHCI 07]


Over 5 months
Exploratory analysis of




Using MatrixExplorer



332 conferences
27 000 actors
118 000 relations
Exploring large networks
Presenting the results
Then MatLink and NodeTrix
Nathalie.Henry@lri.fr
Exploring social networks with matrices
41
Communicating information
InfoVis coauthorship
Nathalie.Henry@lri.fr
AVI coauthorship
Exploring social networks with matrices
42
Communicating information (2)
UIST coauthorship poster
Nathalie.Henry@lri.fr
Exploring social networks with matrices
43
What I have done
Exploration
Perception
1. Make
matrices usable
3. Augment
matrices
2. Combine
matrix+node-link
4. Merge
matrix+node-link
Communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
44
Research directions

Time-related data

On social networks



On other data



Scientific collaboration (INRIA)
Group awareness (Wikipedia France)
Biological networks (Institut Pasteur)
Reflecting on user activity (MSR)
Longitudinal and qualitative studies


Logging data
New visualizations for


analyzing the data collected
reflecting it to users: for exploration and communication
Nathalie.Henry@lri.fr
Exploring social networks with matrices
45
La FIN!
I’m here!

Research approach



Participatory design
Several perspectives
Research directions


Time-related data
Evaluation
Nathalie.Henry@lri.fr
INRIA in 2006
156 teams,
8400 persons
Exploring social networks with matrices
46