Slides of the tutorial "Computational Methods and Tools for Social Network Analysis Networked Learning Communities" at the LAK 2013 in Leuven.
Tutorial Homepage:
http://snatutoriallak2013.ku.de/index.php/SNA_tutorial_at_LAK_2013
Conference Homepage:
http://lakconference2013.wordpress.com/
Presentation by Andreas Schleicher Tackling the School Absenteeism Crisis 30 ...
LAK13 Tutorial Social Network Analysis 4 Learning Analytics
1. Computational Methods
and Tools for Social
Network Analysis of
Networked Learning
Communities
Tutorial at LAK 2013, 9/4/2013
Andreas Harrer, Tilman Göhnert,
Alejandra Martínez-Monés &
Christophe Reffay
2. Agenda
13.30 Introduction of presenters and participants
13.45 Use Cases of SNA4LA
14.15 SNA Basics
15.15 Description of the practical workbench
15.30 COFFEE BREAK
16.00 Hands-on experiences using the workbench
17.30 End of the tutorial
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
2
3. Introduction of presenters
& participants
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
3
4. Tutorial Presenters
• Andreas Harrer
• Tillman Göhnert
• Alejandra Martínez-Monés
• Christophe Reffay
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
4
5. Use Cases of SNA for LA
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
5
6. Identifying Participatory
Roles in CSCL scenarios
Marcos García, J.A., Martínez Monés, A., Dimitriadis, Y., Anguita
Martínez, R. A role-based approach for the support of
collaborative learning activities e-Service Journal. 6(1):40-58,
Diciembre 2007
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
6
7. Participatory roles
• Goal: Identifying roles based on their
position within a network of relationships
o Description of expected roles, based on centrality indexes
o Identify the emergence of those roles in an experience
o Provide them with information adapted to their needs
• Approach:
o Description of roles based on “fuzzy” combinatios of SNA
indexes
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
7
9. Role: Dynamizer student
Indicators
Outdegree CDo(i)
Description Number of links initiated by this actor.
V a l u e s /
Interpretation
A high value, indicates a high participation of
the actor
R e l e v a n c e
rank
First
Outdegree sessions
Description Specifies the relation between participation
and number of sessions
V a l u e s /
Interpretation
A high value indicates a high participation of
the actor in the overall activity
R e l e v a n c e
rank
Second
Indegree CDi(i)
Description Number of links terminating by this actor
V a l u e s /
Interpretation
A medium value indicates a medium
relevance
R e l e v a n c e
rank
Third.
Participatory roles
Dynamizer student
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
9
12. Cohesion in subgroups
Reffay, C. and Chanier, T., (2003) How social network analysis
can help to measure cohesion in collaborative distance
learning, Proc of CSCL, 2003
Reffay, C., Teplovs, C., & Blondel, F.-M. (2011). Productive re-
use of CSCL data and analytic tools to provide a new
perspective on group cohesion. Proc of CSCL 2011.
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
12
13. Cohesion
• Simugline data set
o 4 online groups working on an French as foreign language simulation
o Each group had an instructor and a
• Data
o Discussion forums that are local to each of the 4 groups
• Network
o The relation between “a” and “b” represents messages sent by “a” and
opened by “b” plus messages posted by “b” and opened by “a”
• Indexes
o Cliques at level “c”: subgroup in which the ties between all pairs of agents
have values c or greater (i.e., have exchanged c or more messages).
o “c” can be a value announced by the teacher as the desirable level of
interaction
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
13
14. Comparing groups with (level
10) cliques
Aquitania
Gallia
Lugdunensis
Gallia
Narbonensis
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
14
15. Hierarchical Clusters
GALLIA G
G G G G G G G G G l
l l n l G l n l l l 1
Level 3 2 1 1 t 4 2 6 5 9 0
----- - - - - - - - - - - -
167 . . . XXX . . . . . .
108 . . . XXXXX . . . . .
83 . . XXXXXXX . . . . .
64 . . XXXXXXXXX . . . .
52 . XXXXXXXXXXX . . . .
42 XXXXXXXXXXXXX . . . .
29 XXXXXXXXXXXXXXX . . .
9 XXXXXXXXXXXXXXX XXXXX
5 XXXXXXXXXXXXXXXXXXXXX
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
15
16. PaNern (Star) => Intensity?
Aquitania
Lugdunensis Narbonensis
GalliaIntensity=192
Intensity=12 Intensity=72
Intensity=111
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
16
17. The “fourth” man
Malzahn, N., Harrer, A., & Zeini, S. (2007). The Fourth Man -
Supporting self-organizing group formation in learning
communities. In Proc. of CSCL 2007 (pp. 547–550).
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
17
18. 18
Person-‐‑
Topic-‐‑
Network
from Forum:
group
searches for
the „fourth
man“
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
19. 19
Network
using
semantic
relations
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
20. Blockmodeling
Harrer, A. & Schmidt, A. (to appear 2013). Blockmodeling and
role analysis in multi-relational networks. Social Networks and
Mining. Springer. 2013
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
20
21. 21
Complex networks – dissolving the Death Star
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
22. 22
Complex networks – dissolving the Death Star
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
23. 23
Complex networks – dissolving the Death Star
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
24. A Blockmodel of this network –
positions and reduced matrix
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
24
26. SNA basics
• What is a Social Network?
• Types of networks and network transformations
• Useful definitions and measures on graphs
• Grouping concepts
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
26
27. What is a social network?
• A set of nodes (actors)
o Persons
o Groups
o Organizations
o Objects
o …
• A set of relationships
o Is a friend of
o Is neighbour of
o Provides goods to …
o Has sent a message to …
o Etc.
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
27
28. What is a social network?
• Complexity may
increase.
• Analysis cannot
be done by
hand
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
28
29. Ego-‐‑net : The network of
ego
• Ego: the selected node
• Alters (neighbours): distance (Ego,Alter) ≤ 1
o Ties between ego and alter
o Ties between alters
Whole network Ego-net (x34)Ego-net (x38)
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
29
30. Types of Social Networks
According to…
• Number of sets of actors
o One-mode : one set of actors
o Two-mode : (Bi-partite, affiliation networks) two sets of actors
• Relationships
o Directed or undirected
o Valued or un-valued (1/0)
• How are they built
o Complete networks
o Ego-networks
9/04/13Computational Methods and Tools for Social Network Analysis
of Networked Learning Communities 3030
31. One-‐‑mode or two-‐‑mode
networks
All nodes are of the same type
• Administrators
• Societies
Two-modeOne-mode
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
31
Nodes belong to two
sets
• Students
32. Directed vs Undirected
graphs
• Directed
Undirected
Edges are oriented
Edges are not oriented
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
32
33. Weighted (valued) vs
Unvalued graphs
• Weighted/Valued • Unvalued
Edges have values
Edges have no value
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
33
35. Network types
transformation allowed
Two-‐‑Mode
One-‐‑Mode
Directed
Undirected
Valued
Unvalued
More information
Less Information
Selection strategy
Not reversible!
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
35
36. Two-‐‑Mode
One-‐‑Mode
2
2
1
1
1
1
Do blue nodes share any orange resource? => Unvalued
How many orange resource do blue nodes share ? => Valued
Strategy: Decide what sharing resource represent for relationships between (blue) nodes.
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
36
37. Directed
Undirected
Are nodes connected (one tie is enough)?
Are nodes connected with reciprocal edges?
Strategy: Decide if you have/not edges in both directions.
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
37
38. Valued
Unvalued
Threshold=5
Strategy: Only ties with value>=Threshold are considered
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
38
39. Useful measures of social
networks
• Density
• Degree, In-degree, Out-degree
• Path, Geodesic distance, Diameter
• Centrality indexes (for nodes)
o Degree centrality
o Betweenness centrality,
o Closeness centrality
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
39
41. Density (of edges) for a
directed graph
Eff.=0
Poss.=20
d=0
Eff.=4
Poss.=20
d=0.2
Eff.=8
Poss.=20
d=0.4
Eff.=16
Poss.=20
d=0.8
Eff.=20
Poss.=20
d=1
Reciprocal edges count twice
(twice more possible edges)
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
41
42. 1 4
2 6
5 8
3 7
Net A
1
2
4 5
6
8
3 7
Net B
The structure as a constraint
Do nodes “4” and “5” have the same role in nets A and B?
Density:
DA=9/28=0,321
Density:
DB=9/28=0,321
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
42
43. Centrality
• Who is central in this
network?
439/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
44. Degree in an undirected
graph
• For a node, Degree = number of edges
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
44
45. In-‐‑ & Out-‐‑ degree in an
directed graph
In-‐‑degree = number of edges coming into the node
Out-‐‑degree = number of edges coming out of the node
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
45
46. Path : sequence of edges
connecting 2 nodes
A
H
B
C G
D
E F I J
From A->E : 2 possible paths:
• (A B C E)
or
• (A B D E)
Example in a directed graph
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
46
47. Path: example in an
undirected graph
A
H
B
C G
D
E F I J
From A->E : 2 possible paths:
• (D E)
or
• (D B C E)
Geodesic Distance:
Length of the shortest path
d(D,E) = 1
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
47
48. Diameter of the graph
• Diameter = longest distance in the graph
= maximal distance between any pair of nodes
What is the diameter of this graph?
D = 7
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
48
49. Betweenness centrality
• Number of shortest paths passing through the node
Directed graph
Undirected graph
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
49
50. Closeness centrality
Scoring the closeness of one node to all others
Undirected graph
Directed graph
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
50
51. The Moreno’s
experiments (1943)
Pupils relation in the
classroom:
• Pupils of various age
range
• Gender study
« If you could choose freely,
which are the (2) kids you
would like to have as
direct neighbour? »
Main results:
At <age> => pupils tend to <?>
• 6-8 years old => mix
• 8-13 years old => separate
• 13-15 years old => mix
• 15-17 years old => separate
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
51
52. Moreno’s network
• Who is central in this network?
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
52
53. Components
Removing bridges
(cut-‐‑points)…
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
53
54. …This results in breaking
the component
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
54
55. Grouping concepts – an
overview
Groups can be determined according to different criteria
• Reachability and Distance – group member is
connected via short ways to all other group members
o Direct links – Clique as complete subgraph
o Relaxing the distance – n-Clique requires all nodes being connected via short
path (lesser and equal than n)
• Node degree – group member should be connected to
many group members
o Leaving out a small number of group members: k-Plex
o Having at least k group members as direct neighbours – k-Core
• Contrasting “ingroup” and “outgroup” – density inside is
much higher than outside
o Alliance: only links to ingroup, no links to outside
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
55
56. Grouping concepts – an
overview
• Group concepts fall in two categories:
o Overlapping concepts
• e.g. Cliques
o Disjunct concepts
• e.g. k-cyclic blocks
• Depending on the type of analysis both categories
have their merits
o Disjunct concepts allow clear-cut assignment to one group
o Overlapping concepts allow analysis of transfer ideas, e.g. Clique
percolation
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
56
57. Cliques or K-‐‑cliques
Clique: maximum subset where all
nodes are connected
K-clique: Clique with K members
How many
cliques?
• One 5-clique
• One 4-clique
• One 3-clique
• Three 2-cliques
=> 6 cliques
Which are… ?
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
57
58. K-‐‑cores
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
58
Taken from: V. Batagelj, A. Mrvar / Social Networks 22 (2000) 173-186
59. Clique Percolation
Method
• CPM allows overlapping communities
• Idea: a k-clique “percolates” through the graph
• Overlapping members can be “brokers” between
groups
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
59
Taken from: Wikipedia
61. Practical Workbench
Presentation
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
61
62. Task one: Simuligne
Data Preprocessing
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
62
Raw Data: A network
with weighted, directed
edges
(Number of forum posts
opened)
Preprocessing:
Symmetrisation of edge
weights
(by minimum, maximum,
sum, or average)
63. Task one: Simuligne
• Choose the data set based on preprocessing
o Narbo_Max: Maximum of both directions
o Narbo_Mean: Average of both directions
o Narbo_Min: Minimum of both directions
o Narbo_Sum: Sum of both directions
• Think of the format transformation (UCINET -> SISOB)
• Focus on the appropriate intensity level of the
relation
• Identify groups
• Choose an appropriate output representation
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
63
64. Task two: Collaboration
over Artifacts (BSCW)
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
64
• Two node types
o BSCW (document sharing) folders as artifacts (-..)
• One folder for general information
• Folders for individual case studies
o Pairs of students, each working mainly on a single case (x..)
• Edges weighted by access
65. Task two: Collaboration
over Artifacts (BSCW)
9/04/13
Computational Methods and Tools for Social Network
Analysis of Networked Learning Communities
65
• Choose one of the data sets
o sp1_B_cli_cp_U.txt
o sp2_B_cli_cp_U.txt
o sp3_B_cli_cp_U.txt
o spf_B_cli_cp_U.txt
• Think of the format transformation (UCINET -> SISOB)
• Try to identify the general folder
• Try to identify the projects the pairs of students
worked on
• Analyse the collaboration between the students
(hint: Folding is also in the R-Analysis)