Complejidad y Redes

Network models. Random Networks

Complejidad y Redes.
Universidad Politécnica de Madrid

Designed by starline / Freepik

 Overview

Introduction

The Random graph model

The Small world property The Small world model

Random graphs with general degree distributions

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 2
 Slides based on:

Network Science
slides by L. Barabási and R. Sinatra
Chapter 3 Original slides available at: http://barabasi.com/networksciencebook/

Nuevas Tecnologías y Empresa

slides by J.I. Santos
Chapter 6 and 7 https://sites.google.com/site/meetnachosantos/

Networks: An Introduction.
Book by M.J. Newman

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 3
 Introduction

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 4
 Why models?

So far, we have studied how to measure the properties of a network.

If we want to generalize and make inference about networks with the same properties, we need
models.

“In general, a random graph” is a model network in which some specific set of parameters take fixed
values, but the network is random in other respects” (Newman, 2010).

Random network models: abstractions that reproduce some of the properties of complex networks,
either for:
• Inferring information about the nature of those properties.
• Inferring information about the possible mechanisms of network formation.
• Be a substrate to study the dynamics of phenomena in networks.

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 5
 Random Network Theory

A stochastic model defines a random process of formation of links and nodes.

It defines a set (ensemble) of networks that can appear with a certain probability, i.e., a
probability distribution over the set of all possible networks.

Consequently, the statistical properties of the model should not be confused with the
particular properties of a particular realization.

The idea is to study the average properties of ensembles of graphs.

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 6
® Translated slide from NNTT y Empresa, by J.I. Santos
 Random Networks

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 7
Random graph G(n, m) The Erdős–Rényi model

“One of the simplest examples of a random graph is the network in which we fix only the
number of nodes n and the number of links m.

That is, we take n nodes and place m links among them at random.

More precisely, we choose m pairs of nodes uniformly at random from all possible pairs and
connect them with a link. Typically, one stipulates that the network should be a simple
graph, i.e., that it should have no multilinks or self-links.”

“Erdös and Rényi used this definition in their string of papers

on random networks." Network Science by A.L. Barabási

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 8
® From Networks: An Introduction, by M.J. Newman
Example: G(n,m) = G(10,20), 3 replications

#nodes: 10 #nodes: 10 #nodes: 10

#links: 20 #links: 20 #links: 20
Average degree: 4.0000 Average degree: 4.0000 Average degree: 4.0000
Max degree: 7 Max degree: 7 Max degree: 7
Min degree: 2 Min degree: 2 Average Min degree: 3 Average
Average Clustering: 0.5500 Clustering: 0.4814 Clustering: 0.3619

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 9
Random graph G(n, m) The Erdős–Rényi model

“The random graph model is not defined in terms of a single randomly generated network,
but as an ensemble of networks, i.e., a probability distribution over possible networks.
Thus, the model G(n, m) is correctly defined as a probability distribution P(G) over all graphs
G in which P(G) = 1/Ω”

“When one talks about the properties of random graphs one typically means the average
properties of the ensemble”

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 10
® From Networks: An Introduction, by M.J. Newman
Random graph G(n, m) The Erdős–Rényi model

$ $($'()
The maximum number of links with n nodes: 𝐿!"# = % =
%

The number of ways we can select m links out of 𝐿!"# possible ones is

𝐿!"# 𝐿!"# !
𝐶*!!"# = =
𝑚 𝐿!"# − 𝑚 ! 𝑚!

So 𝑃𝑟+ = (𝐶*!!"# )'(

Some properties of the random graph G(n, m) are straightforward to calculate; obviously, the average
number of link is m, for instance, and the average degree is <k> = 2m/n.

Unfortunately, other properties are not so easy to calculate, and most mathematical work has actually
been conducted on a slightly different model that is considerably easier to handle. This model is called
G(n, p).

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 11
® From Networks: An Introduction, by M.J. Newman
Example: G(10,20), 3 replications

n=10
𝑛 𝑛(𝑛 − 1)
𝐿$%& = = = 45
2 2
𝐿$%& 𝐿$%& !
𝐶'$!"# = = = 3.17 ∗ 10()
𝑚 𝐿$%& − 𝑚 ! 𝑚!

𝑃𝑟* = (𝐶'$!"# )+( = 3.15 ∗ 10+(,

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 12
Random graph G(n, p) The Erdős–Rényi-Gilbert model

“The random network model G(n, p) was independently introduced by Edgar Nelson Gilbert
(1923-2013) the same year Erdős and Rényi published their first paper on the subject.”
Network Science by A.L. Barabási

In G(n, p) we fix not the number but the probability of link between nodes. Again, we have
n nodes, but now we place a link between each distinct pair with independent probability p.

Again, the technical definition of the random graph is not in terms of a single network, but
in terms of an ensemble, a probability distribution over all possible networks. To be specific,
G(n, p) is the ensemble of networks with n nodes in which each simple graph G appears
with probability
!
𝑃𝑟! = 𝑝" (1 − 𝑝) " #"

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 13
® From Networks: An Introduction, by M.J. Newman
Random graph G(n, p)

The number of graphs with exactly n nodes and m links is equal to the number of ways of
picking the positions of the links from the distinct node pairs. Each of these graphs appears
with the same probability PrG, and hence the total probability of drawing a graph with m
links from our ensemble is:
$
!
% " #"
𝑃 𝑚 = 𝑝" (1 − 𝑝)
𝑚

which is just the standard binomial distribution. Then the average number of links is
!
"
𝑛
𝑚 = * 𝑚𝑃 𝑚 = 𝑝
2
"&'

So, <m> is the expected number of links in our model network.

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 14
® From Networks: An Introduction, by M.J. Newman
Random graph G(n, p)

The average degree in a graph with exactly m links is <k> = 2m/n, and hence the average
degree in G(n, p) is
!
"
2𝑚 2 𝑛
𝑘 = * 𝑃 𝑚 = 𝑝= 𝑛−1 𝑝
𝑛 𝑛 2
"&'

The probability that an individual node has k links follows a Binomial (n, p)

𝑛−1 (
𝑝 𝑘 = 𝑝 (1 − 𝑝)$#(#)
𝑘

In other words, G(n, p) has a binomial degree distribution.

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 15
® From Networks: An Introduction, by M.J. Newman
Example: G(n,p) = G(10,0.5), 3 replications
𝑃𝑟! = 𝑝" (1 − 𝑝)*#$%#"

#nodes: 10 #nodes: 10 #nodes: 10

#links: 24 #links: 25 #links: 22
Average degree: 4.8000 Average degree: 5.0000 Average degree: 4.4000
Max degree: 7 Max degree: 8 Max degree: 6
Min degree: 3 Average Min degree: 3 Average Min degree: 0 Average
Clustering: 0.5910 Clustering: 0.5562 Clustering: 0.5800
𝑃𝑟! = 0.5"# (0.5)#$%"# = 2.84 ∗ 10%&# 𝑃𝑟! = 0.5"$ (0.5)#$%"$ = 2.84 ∗ 10%&# 𝑃𝑟! = 0.5"" (0.5)#$%"" = 2.84 ∗ 10%&#

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 16
Random graph G(n, p)

In many cases we are interested in the properties of large networks, so that n can be
assumed to be large (𝑛 → ∞).

Many networks have an average degree that is approximately constant as the network size
becomes large.

In the large-n limit

is the Poisson distribution: in the limit of large n, G(n, p)

has a Poisson degree distribution.

This is the origin of the name Poisson random graph.

® Pict by Network Science by A.L. Barabási

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 17
® From Networks: An Introduction, by M.J. Newman
Clustering coefficient of a random graph

To calculate Ci, we used the expected number of links Li between the node’s ki neighbors:

𝑘+ 𝑘+ (𝑘+ − 1)
𝐿+ = 𝑝 =𝑝
2 2

2𝑒+ 2 𝐿+
𝐶+ = = =𝑝
𝑘+ (𝑘+ − 1) 𝑘+ (𝑘+ − 1)

We also know that 𝑘 = n − 1 p

( (
So: 𝐶+ = ≃
($#)) $

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 18
® Slide from Network Science by A.L. Barabási
Evolution of a random graph G(n, p)
If we vary p from 0 to 1, a phase transition appears with the emergence of a giant
component (component whose size grows in proportion to n).

® Video from Network Science by A.L. Barabási

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 19
Evolution of a random graph G(n, p)

𝑝 < 𝑝! 𝑝! = 1/𝑁

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 20
N Are real networks like random graphs?
7 Subcritical Supercritical Fully Connected
1
Internet
04
Power Grid
27
Science
Collaboration

Actor Network

Yeast Protein
Interactions
and
1 10 <k>

Image 3.8
Real networks (from the perspective of the random model) seem to be mostly in the supercritical
Most real networks are supercritical.
regime
ava- The four (a regimes
giant component, but fragmentation of disconnected components). However, this
predicted by random network theory, marking with a
s es- fragmentation does
cross the location not appear
of several in realofnetworks,
real networks showing
Table 3.1. The diagramthat
indi- the random model does not capture
ain- ancates
important
that mostpart of itsare
networks structure.
in the supercritical regime, hence they are
expected to be broken into numerous isolated components. Only the actor
therComplejidad
network is in the connected regime, meaning that all nodes are expected
y Redes.

cule® Slidetofrom
Universidad Politécnica de Madrid
beNetwork
part of a single
Science by A.L. giant
Network models. Random Networks
Barabásicomponent. Note that while the boundary be-
21
Are real networks like random graphs?

… Random networks do not have hubs. Most nodes have a similar degree.

… p(k) real networks

The degree distribution of a random Erdös-Rényi network does not capture the distributions
of the real networks

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 22
® Translated slide from NNTT y Empresa, by J.I. Santos
Are real networks like random graphs?

If we plot <C>/<k> as a
function of N for several
undirected networks, we find
that <C>/<k> does not
decrease as 1/N, but it is
largely independent of N.

Crand underestimates with

orders of magnitudes the
clustering coefficient of real
networks.
(
𝐶+ =
.

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 23
® Translated slide from NNTT y Empresa, by J.I. Santos
 Are real networks like random graphs?

Average path length

Real networks have short distances

like random graphs.
𝑑 𝐿𝑛 𝑘

Ln 𝑁
𝑑~
Ln 𝑘

m odel?
New

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 24
® Slide from Network Science by A.L. Barabási
Ejemplo

Red de la Red del ejercicio

G(1000, 0.05)
clase 2022 de centralities

𝑘 = 3.04 Supercritical 𝑘 = 9.62 Connected 𝑘 = 49.85 Connected

regime regime regime
𝐿𝑛 𝑁 = 3.91 𝐿𝑛 𝑁 = 7.03 𝐿𝑛 𝑁 = 6.91
𝑘 𝑘 𝑘
𝐶 = 0.81 = 0.0608 𝐶 = 0.254 = 0.0085 𝐶 = 0.049 = 0.0069
𝑁 𝑁 𝑁
𝐿𝑛(𝑁) 𝐿𝑛(𝑁)
𝑑 = 1.59 = 3.52 𝑑 = 3.61 = 3.11 𝑑 = 2.03
𝐿𝑛( 𝑘 ) 𝐿𝑛( 𝑘 )
Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 25
 “Small world” property

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 26
 “Small world” property

Now we look at the "average path length" of a network, that is, what would be the
distance that separates two nodes extracted at random?

Intuition (correct?):
Small networks ~ small distance
Medium networks ~ medium distance
Large networks ~ large distance

How close are the pages of the WWW?

How close are the people in the world? (The world population is approximately 7,753M
people)

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 27
® Translated slide from NNTT y Empresa, by J.I. Santos
 Bacon number

How far is a Hollywood actor from Kevin Bacon? Assume that two actors have a link if they
have shared roles in at least one movie

The oracle of
Bacon

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 28
® Translated slide from NNTT y Empresa, by J.I. Santos
 Erdös number

How far is an investigator from the mathematician Paul Erdös? Assume that two
researchers have a link if they have collaborated on at least one scientific article?

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 29
® Translated slide from NNTT y Empresa, by J.I. Santos
 Stanley Milgram

• A social psychologist
• Yale and Harvard University

• Study on the Small World Problem, beyond

well-defined communities and relations (such
as actors, scientists, …)

• Controversial: The Obedience Study

• What we will discuss today: “An Experimental

Study of the Small World Problem”

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 31
® Slide from Web Science and Web Technology by M. Strohmaier
 The small-world problem

The simplest way of formulating the small-world problem is:

Starting with any two people in the world, what is the likelihood that they will know
each other?

A somewhat more sophisticated formulation, however, takes account of the fact that
while person X and Z may not know each other directly, they may share a mutual
acquaintance - that is, a person who knows both of them. One can then think of an
acquaintance chain with X knowing Y and Y knowing Z.

Moreover, one can imagine circumstances in which X is linked to Z not by a single link, but
by a series of links, X-A-B-C-D…Y-Z. That is to say, person X knows person A who in turn
knows person B, who knows C… who knows Y, who knows Z.

[Milgram 1967, according to ]http://www.ils.unc.edu/dpr/port/socialnetworking/theory_paper.html#2]

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 32
® Slide from Web Science and Web Technology by M. Strohmaier
 An Experimental Study of the Small World Problem [Travers and Milgram 1969]

Milgram’s experiment (1967): send a letter from Nebraska to address Boston using only a
chain of acquaintances

Milgram, S. (1967): The small-world problem. Psychol. Today, 2, 60-67

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 34
® Translated slide from NNTT y Empresa, by J.I. Santos
 Experiment:

Goal
• Define a single target person and a group of starting persons
• Generate an acquaintance chain from each starter to the target

Experimental Set Up
• Each starter receives a document
• was asked to begin moving it by mail toward the target
• Information about the target: name, address, occupation, company, college, year of
graduation, wife’s name and hometown
• Information about relationship (friend/acquaintance)

Constraints
• starter group was only allowed to send the document to people they know and
• was urged to choose the next recipient in a way as to advance the progress of the
document toward the target
Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 35
® Slide from Web Science and Web Technology by M. Strohmaier
 Experiment:

Questions

• How many of the starters would be able to establish contact with the target?

• How many intermediaries would be required to link starters with the target?

• What form would the distribution of chain lengths take?

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 36
® Slide from Web Science and Web Technology by M. Strohmaier
 An Experimental Study of the Small World Problem [Travers and Milgram 1969]

Set Up
• Target person:
• A Boston stockbroker

• Three starting populations

• 100 “Nebraska stockholders”
• 96 “Nebraska random”
• 100 “Boston random”

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 37
® Slide from Web Science and Web Technology by M. Strohmaier
 Results

• How many of the starters would be able to establish

contact with the target?
• 64 out of 296 reached the target.

• How many intermediaries would be required to link

starters with the target?
• Well, that depends: the overall mean 5.2 links
• Through hometown: 6.1 links
• Through business: 4.6 links
• Boston group faster than Nebraska groups
• Nebraska stockholders not faster than Nebraska random

• Milgram himself never used the phrase "six degrees of separation”, but it comes from
the name of a play (and latter movie) written by American playwright John Guare from
1990.
Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 38
® Slide from Web Science and Web Technology by M. Strohmaier
 Follow up work (2008) https://arxiv.org/PS_cache/arxiv/pdf/0803/0803.0939v1.pdf

• Horvitz and Leskovec study 2008

• 30 billion conversations among 240 million people of Microsoft Messenger
• Communication graph with 180 million nodes and 1.3 billion undirected links
• Largest social network constructed and analyzed to date (2008)

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 39
® Slide from Web Science and Web Technology by M. Strohmaier
 Follow up work (2008) https://arxiv.org/PS_cache/arxiv/pdf/0803/0803.0939v1.pdf

Approximation of “Degrees of separation”

• Random sample of 1000 nodes

• for each node the shortest paths to all other nodes
was calculated. The average path length is 6.6.
median at 7.
• Result: a random pair of nodes is 6.6 hops apart on
the average, which is half a link longer than the
length reported by Travers/Milgram.
• The 90th percentile (effective diameter (16)) of the
distribution is 7.8. 48% of nodes can be reached
within 6 hops and 78% within 7 hops.
• We find that there are about “7 degrees of
separation” among people. The distribution of distances in the graph of all active
• Long paths exist in the network; we found paths up Microsoft Instant Messenger user accounts, with an link
joining two users if they communicated at least once
to a length of 29. during a month-long observation period

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 40
® Slide from Web Science and Web Technology by M. Strohmaier
 Small World property [Newman 2003] http://epubs.siam.org/doi/pdf/10.1137/S003614450342480

⟨𝑑⟩ average distance (shortest path) between node pairs in a network

“If the number of nodes within a distance r of a

typical central node grows exponentially with r,
𝑦 = 𝑒$
and this is true of many networks (…), then the
value of 𝑑 will increase as log n. “

“Networks are said to show the small-world Average

effect if the value of 𝑑 scales logarithmically
or slower with network size for fixed mean
𝑦 = ln(𝑛)
degree”
𝑦=𝑛
𝑑 ~ log 𝑁

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks # 41
® based on Slide from Web Science and Web Technology by M. Strohmaier
 ALL MODELS ARE WRONG, BUT SOME ARE USEFUL

George Box

Complejidad y Redes.
Universidad Politécnica de Madrid Network models. Random Networks 62
 ¡Gracias!

Complejidad y Redes.
Universidad Politécnica de Madrid

Designed by starline / Freepik