You are on page 1of 6

International Journal of Computer Systems (ISSN: 2394-1065), Volume 03 Issue 01, January, 2016

Available at http://www.ijcsonline.com/

Study of the Behavior of Different Network Models of Complex Networks


Mini Singh Ahuja1, Dr. Jatinder Singh2, Kriti Sharma3
1

Research Scholar, Punjab Technical University, Jalandhar, India


2
Professor, DAV University Jalandhar, India
3
Student, GNDU Regional Campus, Gurdaspur, India

Abstract
Complex Networks are massive in size and their interactions are complex. Therefore, it is difficult to represent a network
partially. To represent a network partially, we need to study the topology and topological properties of the network. The
goal of studying complex networks is not only to examine the underlying structure of complex systems, but also to learn
how we can control them more efficiently. In this paper we are analyzing the properties of different network models of
complex networks with the real dataset.
Keywords: Graph theory, complex networks, scale free networks, small world networks.

I.

INTRODUCTION

Now days real systems have grown in size


tremendously. They contain million of actors and have
different relationships. Complex networks are the powerful
modeling tools which represent most real world systems.
Complex network paradigm is one of the modeling tools
which have spread through several application fields such
as sociology, communication, computer science, biology,
and physics and so on during last decades. Complex
networks can be represented in the form of large graphs
which have large no of nodes and different types of
relationships with non trival properties. These nodes can be
anything: a person, an organization, a computer or a
biological cell. Nodes can have different size or attributes
which represent a property of real system objects. These
graphs can be directed, undirected or weighted. A complex
network has its roots in graph theory. Few examples of
complex networks are Internet maps ( IP, Routers [36] ,
web graphs (hyperlinks between pages) [37], data exchange
(emails)[38] , social networks (facebook, twitter, scientist
collaboration networks)
biological networks (protein
interaction, epidemic networks) etc. Complex networks
have non trival properties so they cannot be explained by
uniform random, regular or complete models [20]. This has
resulted in definition of set of statistics which have become
fundamental properties of complex networks. These
properties are now being used by many researchers for
studying various phenomenas like spreading of
information [18], protocol performance etc.

II.

ANALYSIS OF COMPLEX NETWORKED SYSTEMS

The complex networked systems should be analyzed to


develop capabilities for understanding, designing and to
control these networked systems. For analyzing these
systems, there are four major elements: understanding
network structure; understanding network dynamics;

mathematical modeling and simulation; situational


awareness, design and control of complex networks.
A. Understanding Network Structure
The underlying topology of a network should be known
beforehand to understand the networks structure and its
behavior. However, the network systems that are of
interest are massive in size and complex. The network
structure of these systems cannot be concluded directly. So,
statistical data and machine learning methods are used to
conclude the network structure of these systems. The major
problem is to conclude the network structure of these
systems from the limited set of noisy observation. These
observations are not directly connected, though they may
be indirectly connected. In many cases, these systems are
of very large scale and may be spatially distributed [13].

B. Understanding Network Dynamics


A networked system serves some dynamic processes
that are the principle objects of interest. Networks
Structure is the underlying layer of Networks dynamics
and also provides some abstraction to the user. The
objective of understanding the network dynamics is to
understand the network behavior under a steady state or
some varying state. In the steady state, behavior possessed
by a network remains constant. But, in varying state,
networks behavior also varies [13].

C. Mathematical Modeling and Simulation


If we are given mathematical or statistical model of a
complex network, we can study the models structure and
produce dynamic networks behavior. Multiple levels of
exactness and accuracy are needed while modeling and
simulating a complex network. At the highest resolution
level, simulations reproduce bits and instructions of the real
complex dynamic networked system. At the lowest
resolution level, sophisticated mathematical models are
required to abstract the crucial information about a

54 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 01, January, 2016

Mini Singh Ahuja et al

Study of the Behavior of Different Network Models of Complex Networks

network. Mathematical modeling of a complex network


and simulating its models structure helps in understanding
the networks behavior and its dynamics. Performance
optimization is done in large-scale simulations. Analysis of
stability, errors and various convergence properties are also
done in complex network simulations [13,17].

In 2001, Albert Laszlo Barabasi et al. [5], studied


scale-free networks. Scale-free networks are sufficient in
nature. Many networks in real-world comprises of scalefree properties in nature. They studied deterministic scalefree networks to understand the topology and properties of
these networks.

D. Situational Awareness, Design and Control of


Complex Network
The fundamental goal of modeling and simulating a
complex network is to develop methods for understanding
networks behavior, controlling various events and remove
anomalies from a complex networked system. The progress
of a complex networked system depends on the study of
these research areas. The final step in analysis procedure of
a complex system focuses on awareness or activeness,
designing and controllability of a complex networked
system. This step of analysis focuses on requirement of
online learning in complex networks and the control
methods are used to modify behavior of a network.
Awareness in a complex network allows the network to
change its dynamics according to the situation, various
design and control methods are used to remove errors and
optimization flow in a network [13,15,19].

III.

LITERATURE REVIEW

In 1999, Albert Laszlo Barabasi and Reka Albert [1],


observed the scaling properties of random networks. They
studied that scale-free properties of a network depends on
following two aspects: expanding of network by adding
new vertices and preferential attachment of new nodes in a
well connected nodes. Networks based on these two
aspects produce large networks with scale-free distributions
which generates robust self-organized networks.
In 2000, Reka Albert and Hawoong Jeong et al. [2],
measured degree of tolerance of various networks against
errors. They examined that even if there is malfunction and
local failures in various networks, these networks show a
surprising degree of tolerance against errors. Hence, they
concluded that error tolerance and attack vulnerability are
the most generic properties of complex networks.
In 2000, Albert Laszlo Barabasi and Reka Albert et
al. [3], measured the scale-free properties of random
networks and studied the topology of www. World Wide
Web is a large directed network in which nodes are web
documents and edges are the links between these web
pages. They demonstrated that despite of the random
character of www it shows various universal scale-free
properties.
In 2000, Albert Laszlo Barabasi and Reka Albert [4],
they studied the topology of complex networks; their local
events and universality. They observed that as the network
grows; their properties changes as the no. of nodes
increases, links connecting these nodes increases and
rewiring between two different nodes. They observed that
topology of these network changes as these properties
changes. They proposed a theory to find the scaling
function and exponents of these networks.

In 2002, Zoltan N. Oltvai and Albert Laszlo


Barabasi [6], reviewed the complex networks in real world.
He compared real world complex networks with the
technological networks and examined how these networks
works while transmitting data from one node to other node.
In 2002, I. Farkas and I.Derenyi et al. [7], analyzed
growing networks ranging from social networks to
biological networks and technological networks for the
demonstrations of scale-free nature in various networks.
They introduced sparse random graph models for
categorizing small-world network models.
In 2003, Xiao Fan Wang and Guanrong Chen [8],
examined the topology and the dynamics of various
complex networks. He also introduced and discussed issue
of robustness versus fragility of synchronization in
complex networks. He discussed how these network
models behave differently with each other in different way.
In 2003, Erzsebet Ravasz and Albert Laszlo Barabasi
[9], studied two common properties of real-world complex
networks in nature and society i.e. scale-free nature and
high degree of clustering. They showed that these two
properties are responsible for hierarchical organization of
complex networks. They also demonstrated that when a
network grows from a small group of nodes which is
organized in a hierarchical order to an increasingly large
group, scale-free topology of the network is maintained.
They also found that various real networks which are
organized in a hierarchy follow strict scaling law in
complex evolving systems. He also concluded that scalefree networks are most robust and fragile in nature due to
the presence of hubs in scale-free networks.
In 2003, Albert Laszlo Barabasi and Eric Bonabeau
[10], studied various scale-free networks in real life and
various applications of scale-free networks. They observed
that scale-free networks are robust and more fragile to
attacks. They studied scale-free networks to understand the
topology and properties of these networks. They designed
networks with optimized network properties.
In 2005, M.E.J. Newman [11], studied random graphs
and described its shortcomings. He also described why
these graphs cannot be used in real-life networks. But
random graphs play a very important role in generation of
various other network models which exhibit those
properties so that they can be used in real-world networks.
In 2006, Matthieu Latapy and Clemence Magnien [12],
measured fundamental properties of real-world complex
networks. They studied how properties of networks change
with the size of the network. They examined that these
networks reach a stage where most of the properties
became stable. . They used various real data sets to analyze
their results.
In 2006, Raoul-Artin Memmesheimer and Marc Timme
[13], suggested new perspectives of designing complex
networks. They studied the relations between dynamics and

55 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 01, January, 2016

Mini Singh Ahuja et al

Study of the Behavior of Different Network Models of Complex Networks

structure of complex networks. They explained how to


design networks if restrictions are imposed on such type of
networks and predefining the network connectivity in
detail. They used random networks and power-law random
networks.

= pN(N-1)/2. The average node degree of the graph is (N1)p.

In 2006, Tom Britton and Maria Deijfen et al. [14],


generated four methods for designing complex networks.
These methods were applied to different network models
and concluded that all of these methods show the correct
degree distribution with large network size.
In 2008, Matthieu Latapy and Clemence Magnien [15],
estimated the relevance between observed properties of
complex networks while measuring complex networks.
They studied how the properties evolve as the network
grows. They used various real data sets to observe the
relevance between complex networks.
In 2011, Yang-Yu Liu and Albert Laszlo Barabasi et.
al. [16], used various methods to control the complex
networks. They developed various analytical tools to study
the controllable measure of complex evolving networks.
They also identified the time dependent control of driver
nodes. They applied these tools to driver nodes and found
that the total no. of driver nodes is illustrated by the
networks degree distribution.
In 2012, Cun-Lai Pu and Wen-Jiang Pei [17], studied
the structural controllability of complex networks. They
studied the robustness of complex networks to observe the
error tolerance of complex networks. They studied random
graph and scale-free models to observe the networks
against attacks and failures. They concluded that degreebased attacks are most efficient than random attacks on
network controllability.
IV.

NETWORK MODELS

Complex Network has been modeled by various types


of graph models. Each graph model shows its specific
features. Three mostly used graph models are:
A. Random Graph of Erds-Rnyi
In late 1950s, two mathematicians, Erdos and Renyi
described a network with complex topology by random
graph [1].. In a random graph, each vertex has random no.
of vertices connected to it [4]. Their work had made a
foundation of the random network theory, and intense
studies had been done in past 50-60 years and even today.
Although, studies clearly shows that many real-life
networks are neither regular nor completely random, but
ER random graph model was the only sensible approach
that made scientists thinking about complex networks for
about half a century.

In a random network, graph may be either connected or


disconnected. A random graph is connected with
probability p. If we start decreasing this probability p, a
stage comes after which the random graph becomes
disconnected. The stage at which the graph becomes
disconnected at a certain probability, this specific
probability is called critical probability (pc). In a
disconnected random graph, the link density p is very low
to form a connected graph and it contains isolated clusters
of nodes. Random networks have shortcomings also. Real
networks cannot be demonstrated by Random Graphs.

B. Scale-free graph of Barabsi- Albert


The scale-free algorithm is generated by BarabsiAlbert. Many Real networks can be demonstrated by scalefree graphs like Internet, World Wide Web, citation
network and some social networks. Scale-free graphs
exhibit all the properties of real networks. Scale-free graph
uses preferential attachment method for network growth.
According to preferential attachment; better the node is
known, more is the probability of a new node to attach to
that node. Scale-free network implements Richer gets
richer theory. In scale-free networks, no. of nodes
increases with the span of time. Hence, in scale-free
networks hubs are formed as the network increases.
Therefore, the average hop count of scale-free graphs is
low than other two graph models. The Barabsi- Albert
model is the famous model which generates scale-free
networks by using preferential attachment and growth
mechanisms. Scale-free network starts initially with mnodes and each time a new node is added is to the present
graph. The network becomes bigger and grows with the
span of times. Scale-free networks always follow powerlaw degree distribution. Power-law describes that vast
majority of nodes are not highly connected but a very few
nodes are highly connected. It decays very slowly so we
can find a hub in the system. Because of the hubs in the
system, scale-free networks are robust and fragile to
attacks.

There are two approaches for generating random graph


models. In first approach, random graph model is
represented by Gr(N,L) with N nodes and L links. Initially,
N nodes are disconnected and there are no links between
them. Then, links are added between nodes at random.
Links which are to be added to the network are chosen
from total of N(N-1)/2 potential links.

C. Small-world model of Watts and Strogatz


Small-world model was discovered by Duncan Watts
and Steven Strogatz. In small-world model, only few links
are used to reach from one node to other node. Social
network and telecommunication networks follow the smallworld property. In small-world network, most of the nodes
are not neighbors of each other. In small-world phenomena
two nodes are linked by a mutual node. Watts and Strogatz
model is the very popular model which implements ring
random graph with small-world property. Small-world
model uses shortcuts to reach from one node to other node.
Hence, average path length is relatively less in case of
small-world model rather than in random graph model and
scale-free model.

In second approach, random graph model is represented


by Gp(N,L) with N nodes and L links. The links are added
to the network with probability p. Then, the number of
links in the graph is not definite and probably equal to E[L]

In this paper, we have focused on the dynamic


evolution of network i.e. a network which grows in size
and recorded our results several times to see how these

V.

RESULT AND SIMULATION

56 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 01, January, 2016

Mini Singh Ahuja et al

Study of the Behavior of Different Network Models of Complex Networks

networks work. We have used three main models of


complex networks and five different data sets to observe
how these networks work in different situation. The real
datasets used are Zachary karate club, Word adjacency
network, Les Miserables network, Dolphins social network
and America football network. Our main aim is to
understand various types of networks and how they work
efficiently with different type of data. We have analyzed
and compared the results to produce the final conclusion.
We hope that it gives insight on how various networks
work differently.
We have implemented each model and studied the
distinct properties of each model. We found that the
growth procedure in random graph follows random link
attachment, the growth procedure in scale-free graph
follows preferential link attachment and the growth
procedure in small-world occurs with the generation of
shortcuts in the graph.

Fig.2 (a) Small-world graph of American football


games,
(b) Scale-free graph of American football games

Table 1: Effect on fraction of reciprocal links and


average clustering coefficient with Random graph using
real data set
Network

Zachary
karate
club

Num
ber of
nodes
34

American
football
network

115

Average
node
degree

Fraction
of
reciproc
al links

Average
clustering
coefficien
t

3.63

41.38%

0.412%

4.11

23.71%

0.215%

Dolphins
social
network

62

3.98

25.10%

0.277%

Les
Miserables
network

77

3.91

24.24%

0.164%

4.07

18.30%

0.219%

Word
adjacency
network

112

Fig.3 (a) Small-world graph of Dolphins social


network,
(b) Scale-free graph of Dolphins social

Fig 4 (a) Small-world graph of Les miserables network,


(b) Scale-free graph of Les miserables network

Fig. 1 (a) Small-world graph of Zachary karate club,


(b) Scale-free graph of Zachary karate club

Fig.5- (a) Small-world graph of Word adjacency


network,
(b) Scale-free graph of Word adjacency network

57 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 01, January, 2016

Mini Singh Ahuja et al

Study of the Behavior of Different Network Models of Complex Networks

Real-world applications of small-world graphs are:


road maps between different cities, railway tracks
between different cities, airline routing map.

Observations and Discussion:


Random graph:
The average node degree of random graph model
applied on various networks is approximately
same which implies that random graph model
cannot be applied on real networks. Because, in
real networks some nodes have high connectivity
while on the other hand some nodes have less
degree.
The degree distribution graph of random model
when applied on various network always follow
Poisson distribution.

VI.

Random networks are different from real networks


because in random network, new nodes are
randomly added to a network while in real
networks new nodes are added in preferential
attachment fashion.

[1]

This is the reason why Random graph model is used


just to understand the complex network. There are
no practical applications of random graph model
in real life.

[3]

REFERENCES

[2]

[4]

Scale-free graph:
The degree distribution graph of scale-free model
shows the best fit line to the frequency of degrees
distribution of the nodes.
The degree distribution plots of our scale-free
graphs show that they follow power-law.
The degree distribution plots of our scale-free
graphs shows that very small no. of nodes are
highly connected and most of the nodes have very
less degree. This shows the existence of hubs in
scale-free graphs.

[5]
[6]
[7]

[8]

[9]

Scale-free graphs are very robust and fragile to


attacks because of the existence of hubs in this
type of network.

[10]

Many real-world applications of scale-free graphs


are: Internet, World Wide Web, social networking
sites.

[12]

Small-world graph:

[13]

The small-world graph is also the practical


application of complex networks other than scalefree graph.
Small-world graphs always show less link
connectivity than scale-free graphs because there
is no existence of hubs in small-world graphs. The
resultant networks of the both the models prove
this point.
Small-world model finds shortcut between two
distant nodes. So, it takes least amount of time
while transferring information from one node to
other node.
In small-world graphs, nodes show stronger link
connectivity with neighbors than other nodes.
While, in scale-free graphs nodes show more link
connectivity with highly connected nodes or hubs.

CONCLUSION

In this paper, we have studied complex network using


its various models. We analyzed various models of
complex network, their properties and simulated our results
to show why these models exhibit these properties. We
have recorded our results several times by changing various
attributes and also using five different real data sets. Each
data set is different from each other. Results of various
models show why they exhibit these properties and why
they are different from each other.

[11]

[14]

[15]

[16]
[17]

[18]

[19]
[20]

Albert Laszlo Barabasi and Reka Albert, Emergence of Scaling


in Random Networks, Science, vol. 286 (1999).
Reka Albert, Hawoong Jeong and Albert Laszlo Barabasi,
Error and attack tolerance of complex networks, Nature, vol. 406,
pg. no. 378-381, 27 July (2000).
Albert Laszlo Barabasi and Reka Albert et. al. , Scale-free
characteristics of random networks: the topology of the world-wide
web, Physica A, Science Direct, pg no. 69-77 (2000).
Albert Laszlo Barabasi and Reka Albert, Topology of Evolving
networks: Local Events and Universality, The American Physical
Society, vol. 85, no. 24, 11 December (2000).
Albert Laszlo Barabasi et al. , Deterministic scale-free
networks, Physica A, Science Direct, pg no. 559-564 (2001).
Zoltan N. Oltvai and Albert Laszlo Barabasi, Lifes complexity
pyramid, Science, vol. 298, pg. no. 763-764 (2002).
I. Farkas and I.Derenyi et al., Networks in life: scaling properties
and eigenvalue spectra, Physica A, Elsevier, vol. 314, pg no. 25-34
(2002).
Xiao Fan Wang and Guanrong Chen, Complex Networks: smallworld, scale-free and beyond, IEEE Circuits and Systems
Magazine (2003).
Erzsebet Ravasz and Albert Laszlo Barabasi, Hierarchical
organization in complex networks, Physical Review E, vol. 67
(2003).
Albert Laszlo Barabasi and Eric Bonabeau, Scale-free
Networks, Scientific American (2003).
M.E.J. Newman, Random graphs as models of networks, in Arxiv
(2005).
Matthieu Latapy and Clemence Magnien, Measuring fundamental
properties of real-world complex networks, Cornell University
(2006).
Raoul-Artin Memmesheimer and Marc Timme, Designing
complex networks, Physica D, Science Direct, vol. no. 224 (2006).
Tom Britton and Maria Deijfen et al. , Generating simple random
graphs with prescribed degree distribution, Journal of Statistical
Physics, Vol. 124, No. 6 (2006).
Matthieu Latapy and Clemence Magnien, Complex Network
Measurements: Estimating the relevance of observed properties, in
infocom (2008).
Yang-Yu Liu and Albert Laszlo Barabasi et. al. , Controllability
of Complex networks, Nature vol. 473, vol. 473 (2011).
Cun - Lai Pu, Wen-Jiang Pei, Andrew Michaelson, Robustness
analysis of network controllability, in Science Direct, pg no..
4420-4425 (2012).
Pastor-Satorras, R. and A. Vespignani. Epidemic spreading in
scale-free networks [Journal Article]. Physical review letters, 2001.
86(14): p. 3200-3203.
Albert Laszlo Barabasi, Network Science, Phil. Transactions of
Royal Society A, vol. 371 (2013).
R. Albert and A. Barab_asi, Statistical mechanics of complex
networks,
Reviews
of
Modern
Physics,
vol.
74,no. 1, pp. 47{97, 2002

58 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 01, January, 2016

Mini Singh Ahuja et al

Study of the Behavior of Different Network Models of Complex Networks

[21] Steven H. Strogatz, Exploring Complex Networks, Nature, vol.


410 (2001).
[22]
Hartwell, L. H., Hopfield, J. J., Leibler, S. & Murray, From
molecular to modular cell biology, Nature, vol. 402, pg. no. C47
C52 (1999).
[23] Jeong H., Tombor, B., Albert, R., Oltavi, Z, N. & A.-L. Barabsi,
The large-scale organization of metabolic networks, Nature, vol.
407, pg. no. 651654 (2000).
[24] Faloutsos, M., Faloutsos, P. & Faloutsos, C., On power-law
relationships of the internet topology, Comp. Comm. Rev., vol. 29,
pg. no. 251262 (1999).
[25] Faloutsos, M., Faloutsos, P. & Faloutsos, C., On power-law
relationships of the internet topology, Comp. Comm. Rev., vol. 29,
pg. no. 251262 (1999).
[26] M. E. J. Newman, The structure of scientific collaboration
networks, Proc. Natl Acad. Sci. USA, vol. 98, pg no. 404409
(2001).
[27] M. E. J. Newman, Models of the small world: a review J. Stat.
Phys., vol. 101, pg. no. 819841 (2000).
[28] Krapivsky, P. L., Redner, S. & Leyvraz, F., Connectivity of
growing random networks, Phys. Rev. Lett., vol. 85, pg. no. 4629
4632 (2000).
[29] Cohen, R., Erez, K., ben-Avraham, D. & Havlin, S., Resilience of
the Internet to random breakdowns Phys. Rev. Lett., vol. 85, pg.
no. 46264628 (2000).
[30] Molloy, M. & Reed, B., The size of the giant component of a
random graph with given degree sequence, Combinatorics
Probab. Comput., vol. 7, pg. no. 295305 (1998).
[31] M. E. J. Newman, Models of the small world: a review, J. Stat.
Phys, vol. 101, pg. no. 819841 (2000).
[32] M. E. J. Newman, Moore, C. & Watts, D. J., Mean-field solution
of the small-world network model, Phys. Rev. Lett., vol. 84, pg.
no. 32013204 (2000).
[33] Kleinberg, J. M., Navigation in a small world, Nature, vol. 406,
pg. no. 845 (2000).
[34] Milgram, S., The small world problem, Psychol. Today, vol. 2,
pg. no. 6067 (1967).
[35] Amaral, L. A. N., Scala, A., Barthlmy, M. & Stanley, H. E.,
Classes of behavior of small-world networks, Proc. Natl Acad.
Sci. USA, vol. 97, pg. no. 1114911152 (2000).
[36] C. Gkantsidis, M. Mihail, and E. Zegura, Spectral analysis of
internet topologies, in IEEE INFOCOM, 2003.
[37] P. Boldi and S. Vigna, The webgraph framework i: compression
techniques, in WWW, 2004
[38] S. Voulgaris, A.-M. Kermarrec, L. Massoulie, and M. van Steen,
Exploiting semantic proximity in peer-to-peer content searching,
in IEEE FTDCS, 2004
[39] Mini Singh A. et al., Future Prospects in Community Detection,
International Journal of Computer Science Engineering and
Information Technology Research, Vol. 4, Issue 5, ISSN: 22496831 (2014).

59 | International Journal of Computer Systems, ISSN-(2394-1065), Vol. 03, Issue 01, January, 2016

You might also like