Exploring Social Networks with Matrix-Based Representations

1
Exploring Social Networks with Matrix-Based
Representations

• Nathalie Henry
• Jean-Daniel Fekete
• INSITU / AVIZ Lab.
• INRIA / Laboratoire de Recherche en Informatique
• Université de Sydney
• Nathalie.Henry_at_lri.fr, Jean-Daniel.Fekete_at_inria.f
  r

2
The problem

• Using Node-Link diagrams to visualize
• Tree-like
• Small-world
• Almost-complete

http//www.infovis-wiki.net/index.php/Social_Netwo
rk_Generation
3
What social scientists are looking for

• What are the communities?
• How actors are linked within the community?
• How communities are linked?
• Who is central?

4
Proposing a readable representation for dense
graphs

• What are the communities?
• How actors are linked within the community?
• How communities are linked?
• Who is central?

?
Ghoniem et al. 05
5
Matrix Visualization
?
?
6
Matrix vs NodeLink

• Intuitive
• Compact
• More readable for path following
• More effective for small graphs
• More effective for sparse graphs
• Useless without layout
• Node overlapping
• Edge crossing
• Not readable for dense graphs
• Manipulation requires layout computation

• Usable without reordering
• No node overlapping
• No edge crossing
• Readable for dense graphs
• Fast navigation
• Fast manipulation
• Usable interactively
• More readable for some tasks
• Less intuitive
• Use more space
• Weak for path following tasks

-
7
Communicate
Explore
8
Participatory Design

• What Social Science researchers
• Use? (representations, software)
• Analyze? (datasets)
• Do? (tasks, exploration process)
• Want? (aspiration)

http//insitu.lri.fr/nhenry/Workshop.html
9
Needs expressed for an exploratory analysis
system

• Multiple representations
• Interaction instead of parameter tuning
• HenryFekete06
• Overviews
• Connected Components Representation
• Global Information on Graph and Social Networks
• Data, Attributes, SNA actors, relationship,
  degree distribution, diameter, 5 most connected,
  5 less connected, centrality measures.
• Multiples représentations Nœuds-liens
  (moreno30s), Matrices (forsyth40s)
• Layout for node-link, ordering for matrices
• Interactions directly on the network
• Filtering, Clustering (multiples), Aggregation
• Compare, Confront, Annotate

10
Possible solutions

• Better layout/ordering
• MatrixExplorer
• MatLink
• NodeTrix
• TreePlus, Links over Treemap, NetLens, Semantic
  Substrates

• Improve one representation
• Combine both representations
• Augment one representation
• Find hybrid representations
• Find other representations

11
1. Improve one representation

• Layout (Node-Link)
• Order (Matrix)

12
Reorder to understand
Bertin, 1967

• Why?
• Survey in progress
• Interactive techniques
• Algorithms for reordering tables
• Algorithms for graphs linearization

13
Identifying Visual Patterns
14
2. Combine both representations

• MatrixExplorer

15
MatrixExplorer HenryFekete06

• Matrices to explore
• Node-Link diagrams to present findings

16
3. Augment one representation

• MatLink

17
MatLinkHenryFekete07

• Solving the path-related tasks problem for
  matrices
• Augmenting matrices with interactive links

18
MatLink significantly improving matrices

• Controlled experiment
• 3 vis. x 6 datasets x 5 tasks
• Matrix , Node-Link, MatLink
• Data From almost-trees
• To complete-graphs
• Including small-world networks
• Tasks 1. CommonNeighbour,
• 2. ShortestPath,
• 3. MostConnected,
• 4. ArticulationPoint,
• 5. LargestClique

19
4. Find a hybrid representation

• NodeTrix

20
NodeTrixHenry et al.07

• Designed for small-world networks
• Globally sparse
• Locally dense
• Visualizing dense sub-graphs as matrices
• Interact to create, edit and remove the matrices

21
NodeTrix

• VIDEO at http//insitu.lri.fr/nhenry/nodetrix/nod
  etrix.mov

22
NodeTrix the NetVis Nirvana?
?

• Can you see every node?
• Can you count each nodes degree?
• Can follow every link from its source to its
  destination?
• Can you idenfity clusters and outliers?
• Node Labels
• Link Labels (excentric labels?!)
• even clusters labels
• Node Attributes
• Link Attributes
• even clusters attributes
• Directed Graph (links width?!)
• But Its gonna be crowded here !

?
?
?
23
Visual Patterns
Cross Pattern
Block Pattern
Mixte Pattern
24
Visual Patterns
Infovis Coauthorship (133 actors)
25
Using Interaction for Story-telling
26
Future Directions

• Scaling up to very large network...
• the problem of reordering
• Provide usable tools to sociologists...
• the problem of bug fixing
• Navigating and aggregating Zame
• Towards collaborative exploration
• From exploration to story telling

27
La Fin
28
References

• N. Henry, J-D. Fekete, M. Mcguffin. NodeTrix
  Hybrid Representation for Analyzing Social
  Networks, Research Report 6183, INRIA, 2007.
  https//hal.inria.fr/inria-00144496
• N. Henry and J-D. Fekete. MatLink Enhanced
  Matrix Visualization for Analyzing Social
  Networks. In Processding of the eleventh IFIP
  TC13 International Conference on Human-Computer
  Interaction (Interact 2007), September 2007.
  Springer Verlag. 14 pages, to be published.
• N. Henry and J-D. Fekete. MatrixExplorer a
  Dual-Representation System to Explore Social
  Networks. IEEE Transactions on Visualization and
  Computer Graphics (Proceedings Visualization /
  Information Visualization 2006), 12(5)677-684,
  September-October 2006.
• M. Ghoniem, J-D. Fekete and P. Castagliola.
  Readability of Graphs Using Node-Link and
  Matrix-Based Representations Controlled
  Experiment and Statistical Analysis. Information
  Visualization Journal, 4(2)114135, 2005.