Social Network Analysis

Eytan Adar
590AI

Some content from Lada Adamic

Vocabulary Lesson

Relational Tie

Actor parentOf
Person supervisorOf
Group reallyHates (+/-)
Event …
… Dyad
Relation: collection of ties of a specific type
(every parentOf tie)
Vocabulary Lesson
If A likes B and B likes C then A likes C (transitivity)
If A likes B and C likes B then A likes C
…

Triad
Vocabulary Lesson

Social
Network
One mode
Vocabulary Lesson

Social
Network
Two mode
Vocabulary Lesson

Ego-Centered
ego
Network
(egonet)
Describing Networks
• Graph theoretic
– Nodes/edges, what you’d expect
• Sociometric
– Sociomatrix (2D matrix representation)
– Sociogram (the adjacency matrix)
• Algebraic
– ni  nj
– Also what you’d expect
• Basically complimentary
Describing Networks
MIT
Stanford
Describing Networks
• Geodesic
– shortest_path(n,m)
• Diameter
– max(geodesic(n,m)) n,m actors in graph
• Density
– Number of existing edges / All possible edges
• Degree distribution
Types of Networks/Models
• A few quick examples
– Erdős–Rényi
• G(n,M): randomly draw M edges between n nodes
• Does not really model the real world
– Average connectivity on nodes conserved
Types of Networks/Models
• A few quick examples
– Erdős–Rényi
– Small World
• Watts-Strogatz
• Kleinberg lattice model
 Small world experiments then

MA

NE

Milgram’s experiment (1960’s):

Given a target individual and a particular property, pass the message to a person
you correspond with who is “closest” to the target.
Watts-Strogatz Ring Lattice Rewiring
Select a fraction p of
edges
Reposition on of their
endpoints

Add a fraction p of
additional edges leaving
underlying lattice intact

• As in many network generating algorithms

• Disallow self-edges
• Disallow multiple edges
Geographical search

MA

NE
Kleinberg Lattice Model

nodes are placed on a lattice and

connect to nearest neighbors

r
additional links placed with puv ~ d uv

Kleinberg, ‘The Small World Phenomenon, An Algorithmic Perspective’

(Nature 2000)
A little more on degree distribution
• Power-laws, zipf, etc.
Distribution of users among
web sites
CDF of users to sites

Sites ranked by popularity

A little more on degree distribution
• Pareto/Power-law
– Pareto: CDF P[X > x] ~ x-k
– Power-law: PDF P[X = x] ~ x-(k+1) = x-a
– Some recent debate (Aaron Clauset)
• http://arxiv.org/abs/0706.1062
• Zipf
– Frequency versus rank y ~ r-b (small b)
• More info:
– Zipf, Power-laws, and Pareto – a ranking tutorial
(http://www.hpl.hp.com/research/idl/papers/ranking/ranki
ng.html)
Types of Networks/Models
• A few quick examples
– Erdős–Rényi
– Small World
• Watts-Strogatz
• Kleinberg lattice model
– Preferential Attachment
• Generally attributed to Barabási & Albert
Basic BA-model
• Very simple algorithm to implement
– start with an initial set of m0 fully connected nodes
• e.g. m0 = 3

– now add new vertices one by one, each one with exactly m
edges
– each new edge connects to an existing vertex in proportion
to the number of edges that vertex already has →
preferential attachment
Properties of the BA graph
• The distribution is scale free with exponent a = 3 P(k) = 2
m2/k3
• The graph is connected
– Every new vertex is born with a link or several links (depending on
whether m = 1 or m > 1)
– It then connects to an ‘older’ vertex, which itself connected to another
vertex when it was introduced
– And we started from a connected core
• The older are richer
– Nodes accumulate links as time goes on, which gives older nodes an
advantage since newer nodes are going to attach preferentially – and
older nodes have a higher degree to tempt them with than some new
kid on the block
Common Tasks
• Measuring “importance”
– Centrality, prestige
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
Data Collection / Cleaning

Analysis
Past
 Small datasets
Data Collection / Cleaning Pretty explicit connections

Analysis
Understand the properties

Past
Present
 Data Collection / Cleaning

Large datasets
Entity resolution
Implicit connections

Analysis

Understand the properties

Present
Common Tasks
• Measuring “importance”
– Centrality, prestige (incoming links)
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
Centrality Measures
• Degree centrality
– Edges per node (the more, the more important the node)
• Closeness centrality
– How close the node is to every other node
• Betweenness centrality
– How many shortest paths go through the edge node
(communication metaphor)
• Information centrality
– All paths to other nodes weighted by path length
• Bibliometric + Internet style
– PageRank
Tie Strength
• Strength of Weak Ties (Granovetter)
– Granovetter: How often did you see the contact that helped you find the job prior
to the job search
• 16.7 % often (at least once a week)
• 55.6% occasionally (more than once a year but less than twice a week)
• 27.8% rarely – once a year or less
– Weak ties will tend to have different information than we and our close contacts
do

weak ties will tend to have high

beweenness and low transitivity
Common Tasks
• Measuring “importance”
– Centrality, prestige (incoming links)
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
Link Prediction

?
Link Prediction in Social Net Data
• We know things about structure
– Homophily = like likes like or bird of a feather flock
together or similar people group together
– Mutuality
– Triad closure

• Various measures that try to use this

Link Prediction
• Simple metrics
– Only take into
account graph
properties

1

z ( x )  ( y ) log |  ( z ) |

Γ(x) = neighbors of x
Originally: 1 / log(frequency(z))

Liben-Nowell, Kleinberg (CIKM’03)

Link Prediction
• Simple metrics
– Only take into
account graph
properties

  l

l 1
| paths l 
x, y |

Paths of length l (generally 1)

from x to y
weighted variant is the number of
times the two collaborated
Liben-Nowell, Kleinberg (CIKM’03)
Link Prediction in Relational Data
• We know things about structure
– Homophily = like likes like or bird of a feather flock
together or similar people group together
– Mutuality
– Triad closure

• Slightly more interesting problem if we have

relational data on actors and ties
– Move beyond structure
Relationship & Link Prediction

advisorOf?

Employee /contractor
Salary
Time at company
…
Link/Label Prediction in Relational Data
• Koller and co.
– Relational Bayesian Networks
– Relational Markov Networks
• Structure (subgraph templates/cliques)
– Similar context
– Transitivity

• Getoor and co.

– Relationship Identification for Social Network Discovery
• Diehl/Namata/Getoor AAAI’07
– Enron data
• Traffic statistics and content to find supervisory relationships?
– Traffic/Text based
– Not really identification, more like ranking
Common Tasks
• Measuring “importance”
– Centrality, prestige (incoming links)
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
Epidemiological
• Viruses
– Biological, computational
– STDs, needle sharing, etc.
– Mark Handcock at UW
• Blog networks
– Applying SIR models (Info Diffusion Through Blogspace, Gruhl et al.)
• Induce transmission graph, cascade models, simulation
– Link prediction (Tracking Information Epidemics in Blogspace, Adar et
al.)
• Find repeated “likely” infections
– Outbreak detection (Cost-effective Outbreak Detection in Networks,
Leskovec et al.)
• Submodularity
Common Tasks
• Measuring “importance”
– Centrality, prestige (incoming links)
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
D om ingo
C arlos
A lejandro
E duardo
F rank
H al
K arl
B ob
Ike
G ill
Lanny
M ike
John
X avier
U trecht
N orm
R uss
Q uint
W endle
O zzie
Ted
S am
Vern
P aul
Blockmodels
• Actors are portioned into positions
– Rearrange rows/columns
• The sociomatrix is then reduced to a smaller
image
• Hierarchical clustering
– Various distance metrics
• Euclidean, CONvergence of CORrelation (CONCOR)
• Various “fit” metrics
 Im age m atrix
C ohesion C enter-periphery R anking
Girvan-Newman Algorithm
• Split on shortest paths (“weak ties”)

1. Calculate betweenness on all edges

2. Remove highest betweenness edge
3. Recalculate
4. Goto 1
Other solutions
• Min-cut based
• “Voltage” based
• Hierarchical schemes
Common Tasks
• Measuring “importance”
– Centrality, prestige (incoming links)
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
Network motif detection
• How many more motifs of a certain type exist
over a random network

• Started in biological networks

– http://www.weizmann.ac.il/mcb/UriAlon/
Basic idea
• construct many random graphs with the same
number of nodes and edges (same node
degree distribution?)
• count the number of motifs in those graphs
• calculate the Z score: the probability that the
given number of motifs in the real world
network could have occurred by chance
Generating random graphs
• Many models don’t preserve the desired
features
• Have to be careful how we generate
Other Structural Analysis

sisterOf

hte rOf
g
dau
c e Of Find all
e
ni
Common Tasks
• Measuring “importance”
– Centrality, prestige (incoming links)
• Link prediction
• Diffusion modeling
– Epidemiological
• Clustering
– Blockmodeling, Girvan-Newman
• Structure analysis
– Motifs, Isomorphisms, etc.
• Visualization/Privacy/etc.
Privacy
• Emerging interest in anonymizing networks
– Lars Backstrom (WWW’07) demonstrated one of
the first attacks
• How to remove labels while preserving graph
properties?
– While ensuring that labels cannot be reapplied
Network attacks
• Terrorist networks
– How to attack them
– How they might attack us
• Carley at CMU
Software
• Pajek
– http://vlado.fmf.uni-lj.si/pub/networks/pajek/
• UCINET
– http://www.analytictech.com/
• KrackPlot
– http://www.andrew.cmu.edu/user/krack/krackplot.shtml
• GUESS
– http://www.graphexploration.org
• Etc.
Books/Journals/Conferences
• Social Networks/Phs. Rev
• Social Network Analysis (Wasserman + Faust)
• The Development of Social Network Analysis
(Freeman)
• Linked (Barabsi)
• Six Degrees (Watts)
• Sunbelt/ICWSM/KDD/CIKM/NIPS
Questions?
Assortativity
• Social networks are assortative:
– the gregarious people associate with other gregarious people
– the loners associate with other loners
• The Internet is disassorative:
Disassortative:
Random hubs are in the periphery
Assortative:
hubs connect to hubs