An Agent-based Simulation of Extremist Network Formation through

Radical Behavior Diffusion

Carlos Sureda Gutierrez, Benoit Gaudout and Frederic Amblard

University of Toulouse, 2 Rue du Doyen-Gabriel-Marty, 31042 Toulouse, France

Keywords: Network Formation, Cultural Transfer, Social Simulation, Radicalization, Terrorism, Agents.

Abstract: Understanding how terrorist networks are created and how individuals turn into extremism and then into

terrorism is a current subject of interest and a cross-domain research problem since it involves scholars from

political sciences, sociology, physics and computer scientists among others. In this paper, an agent-based

approach is used to simulate the process of radicalization and creation of a terrorist network, and the link

between both processes. Each citizen has several attributes allowing the model to take into account

heterogeneous profiles of individual. Furthermore, we model the social transfer that takes place during the

interaction of individuals in order to understand how cultural ideas are transmitted. This paper also provides

a non-exhaustive but detailed survey of the state of the art on the agent-based terrorist networks modelling.

1 INTRODUCTION

We have seen in the last years an increase of news

related to terrorism issues, and it is nowadays

considered as one of the main global issues. This is

the case of France in particular, where the fight

against the terrorism is the second main

preoccupation of the population (Lévy, et al., 2016),

specially after the terrorist attacks suffered recently.

Nevertheless, extremism principles and its

corresponding practice implementation – terrorism –

have existed long before in another form and with

others goals. For example, in the late 20

century

some regional terrorist groups as Euskadi Ta

Askatasuna (ETA) in Spain or Irish Republican

Army (IRA) in Ireland had the main attention of

regional media.

Terrorist Networks (TN) works can be studied

according to several points of views, methodologies

and goals, doing its classification very complex, and

could even lead to a publication on this subject. A

first look can be taken by type of approach:

quantitative vs. qualitative, even if some authors

support a combined approach (Crossley, 2010).

Quantitative studies try to find a model that explains

the system withouth taking into account the

purposeful action of actors. On another side, the

qualitative approach focuses on the context in which

the relationships are created within the network and

the different types of links. Qualitative studes

(Taylor, 2010) try to explain how the particular

situation of actors at every moment drives the

structure of networks (Hidalgo, 2016). The focus can

also be made on the diffusion of behaviors within a

network, as (Deffuant, et al., 2002) (Holme and

Newman, 2006). Likewise, there are some

contributions on terrorist behavior modelling

(Taylor, 2010).

Most of the work done on TN takes the network

as a macro macro object where the structure cannot

emerge from heterogeneous egos, hence the utility

of an agent-based approach. Agent-based simulation

models make possible to analyse the macroscopic

effects of the heterogeneous agents’ interactions.

There are some papers in the literature that deal with

TN using an agent-based simulation approach.

Among them, different topics that can be analysed:

radicalization (MacKerrow, 2003) (North, et al.,

2004), recruitment (Berry, et al., 2004) (Li, et al.,

2015), operational tasks and roles (O'Neil, 2012) (Li,

et al., 2015), knowledge acquisition (Moon and

Carley, 2007), counter-terrorism (Keller, et al.,

2010) (Ilachinski, 2012).

Nevertheless, none of papers described above

focuses on the simulation of TN formation process

as a subject of study, combining at the same time the

diffusion of radicalism, heterogeneous behaviours

and micro-macro mechanisms modifying the form of

the network. The model presented in this paper deals

236

Sureda C., Gaudou B. and Amblard F.

An Agent-based Simulation of Extremist Network Formation through Radical Behavior Diffusion.

DOI: 10.5220/0006198602360243

In Proceedings of the 9th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2017), pages 236-243

ISBN: 978-989-758-219-6

with these issues. Moreover, cultural transfers and

meeting mechanisms are introduced.

This paper is organized as follows: in section 2

we discuss in detail about agent-based simulation

approaches and propose a set of key aspects when

modelling TN; in section 3 our model is presented

and compared to other papers through these key

aspects; then in section 4 show some results. Finally,

section 5 concludes the paper.

2 AGENT-BASED SIMULATIONS

OF TERRORIST NETWORKS

Since the complexity and variety of agent-based

models of TN, we propose in the next paragraph a

detail of central questions and common aspects that

are studied in the state of the art and discuss about

them. The papers analysed here have been selected

in order to match all the specifications of our

research: and agent-based simulation, the notion of

social network and modelling of terrorism.

Each paper focuses more often in a central

question that represents the main idea (ex. Counter-

terrorism) responding at the same time to some other

aspects that can often be common to other papers

(ex. How are the links created?). These aspects have

been chosen in order to fit with the actual interest in

this paper, that is the network formation.

Thus, the following main questions and common

aspects have been identified.

2.1 Main Questions

2.1.1 Counter Terrorism

One of the most important questions addressed in

TN studies is the development, analyse and

simulation of counter-terrorism strategies.

(Raczynski, 2004) introduces the infiltration of

counter-terrorist agents within the TN, together with

the possibility of some terrorist agents collaborating

with anti-terrorist structures. Therefore, terrorist

agents can be neutralized leading to the destruction

of terror structures. (Ilachinski, 2012) developed an

exhaustive framework taking into account different

dynamics, and it focuses on the co-evolution of both

terrorist and counter-TN, using some measures as

entropy, vulnerability or cohesion. Moreover,

counter-terrorist agents can detect, infiltrate and

capture terrorist agents and/or links. (Tsvetovat and

Carley, 2004) simulates the behaviour of the terrorist

structure under attacks aiming to destabilize the

network. In order to do it, the anti-terrorism agents

have two goals: to learn the structure, tasks and

knowledge distribution of terrorists and to remove

and isolate terrorist agents. Regarding the first goal,

there are 3 ways of intercepting information spread

through the TN (random, snowball and socially

intelligent traffic analysis). Once the counter-

terrorist agents have enough information about the

TN, they can attack a random point, the one with the

highest degree centrality or cognitive load. (Genkin

and Gutfraind, 2011) studies different profiles (lone

and trapped wolfs, wolf and trapped wolf packs) and

scenarios of radicalization through isolation,

clustering and mean cell size measures. Occlusion

and encapsulation mechanisms are explained from a

SNA point of view and presented as counter-

terrorism strategies. (Keller, et al., 2010) analyses 4

counter-terrorism strategies and their corresponding

response from terrorist agents: leader-focused,

grassroots, geographic and random. Terrorist agents

modify their parameters in order to protect

themselves from these attacks. For example, they

increase the cost of creating new links in the case of

a leader-focused anti-terrorist strategy, since having

too many links is problematic. The same strategy is

adopted by terrorist agents in (O'Neil, 2012), where

highly exposed terrorist agents are killed or captured

et each simulation step. Each terrorist agent has a

secrecy attribute, that reflects the agent’s ability to

remain unexposed. Regarding (Li, et al., 2015), the

counter-terrorism aspect is taken into account by

allowing the network to be recovered after an attack.

The TN can be perturbed by eliminating either a

leader or a regular member of a terrorist cell, and the

reconstruction process is explained.

2.1.2 Behaviour Diffusion

Modelling the terrorist’s behaviour and its diffusion

is another approach used for studying TN formation.

Behaviour and opinion are two related concepts that

can raise some confusions. In this paper, an agent

behaviour and opinion is equivalent, and is

completely defined through its characteristic vectors

of characteristics (cf. paragraph 3.2).

(MacKerrow, 2003) focus in social bargaining

process that allows a bottom-up approach for belief

spreading. He also introduces the grievance toward a

social group using socio-economic and cultural-

penetration metrics, and the notions of social welfare

and social capital, as well as a social pressure.

(Berry, et al., 2004) developed a whole framework

to model behaviour diffusion. Agents’

disgruntlement evolves according to its personal

network, society attitude, mosques they are member

An Agent-based Simulation of Extremist Network Formation through Radical Behavior Diffusion

237

of and finally the clique behaviour, that is the

behaviour of agents surrounding it. Disgruntlement

changes following a linear reinforcement and/or

linear attraction. (Genkin and Gutfraind, 2011)

model the pressurability on an agent, that measures

how fast an individual changes his opinions

according to his friends’ ones, an essential notion in

order to model how a moderate agent becomes a

radical one.

2.1.3 Operations, Tasks and Roles

An important part of TN studies is focusing about

the objective of their members. Some papers take

into account the organization of a terrorist attack by

modelling different roles, tasks or necessary

operations in order to perform a mission.

(O'Neil, 2012) and (Li, et al., 2015) develop a

whole framework dedicated to operations, roles and

tasks, with hierarchical structures and leader

selection process. (Ilachinski, 2012) developed a

model of TN as a complex adaptive system, with

very complex roles, missions and dynamics.

(Raczynski, 2004) uses a probability of terrorist

attack based on the size of the TN as well as the

number of existing terrorist-supporting structures.

(Tsvetovat and Carley, 2004) model different tasks

and its assignation to agents through a meta-matrix.

Concerning (Moon and Carley, 2007), the objective

of individuals is knowledge acquisition (a pre-

condition to a terrorist attack).

2.1.4 Recruitment

In order to become a terrorist, a radical agent in

(Berry, et al., 2004) has to belong to a radical clique

and to be contacted by a “bridge” agent, that is an

agent who allows to interact with a terrorist

organization. A similar idea is used by (MacKerrow,

2003), where an agent has to be radicalized before

contacting a terrorist organization. Moreover

(MacKerrow, 2003) proposes a parameter of “self-

motivation toward terrorism” that allow some agents

to require more or less active recruitment. (Li, et al.,

2015) introduces a recruitment cell that

communicates with other cells in order to know

about human resources needs. They define also the

processes of recruitment and a parameter of

recruitment cost. (O'Neil, 2012) take also into

account a recruiter role and proposes a mechanism

of replacement of a leader.

2.2 Common Aspects

In order to answer the previous questions, we have

identified some more general aspects tackled by

authors. These aspects are not the main issue of

respective papers but they are however necessary to

answer some central questions. The first three (Link

creation, Structure and process relation and

Topology) correspond to network formation related

issues, and the last two (Multilayer and In/Out data)

are more methodological.

2.2.1 Link Creation

This is the most important step in a simulation

model of network formation.

(Berry, et al., 2004) calculates the weight of a

relationship between two agents, based on the

disgruntlement level of agents and their personal

traits. A link is created from a certain threshold.

(MacKerrow, 2003) uses a social bargaining

between agents since they seek to get into a leader’s

network, that is someone with a lot of relationships,

so creating a link has a “social cost”. (Raczynski,

2004) creates a link when agents are spatially close,

and this is more likely when they are in cities, an

abstract agent of his model. (North, et al., 2004)

focus on the inference from incomplete network

data, together with some social network rules and

detailed documentation about interactions between

agents. (Ilachinski, 2012) link dynamics is based in

one hand on the theory of communication of (Monge

and Contractor, 2003), that is social reward and

mission gain, and in the other hand in an equation

calculating the perceived risk of creating a link.

Furthermore, agents have some motivations to create

or erase links, as acquiring mission-required

resources or the perceived risk of discovery.

(Tsvetovat and Carley, 2004) and (Moon and Carley,

2007) give a detailed equation to select the best

candidate to create a link to and also a probability of

interaction based on differences between agents’

attributes, social distance and spatial proximity.

(Genkin and Gutfraind, 2011) follows the “never

swap a good friend” rule. This is based on

homophily, magnets (the equivalent of social spots

in this paper), degree budget (maximum of

connections by agent), transitivity and an attrition

parameter measuring the migration rate of radicals

among different communities. (Keller, et al., 2010)

applies the preferential attachment principle.

(O'Neil, 2012) shows a formula calculating the

weight of links, based on agents’ roles and locations.

Moreover, there is a mechanism controlling the

macro behaviour of the network, depending on the

success of terrorist operations. (Li, et al., 2015)

specifies some scenarios where a link can be created.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

238

For example, when an agent transfers a resource

from a cell to another, links created by terrorist

recruitment cells when they capture new terrorist,

and the designation of a new leader.

2.2.2 Structure and Process Relation

Some authors explain how network’s structure and

agents’ behaviours interact, independently of link

creation process.

For example, in (O'Neil, 2012) the network

structure changes according to the issue of

operations. In (Li, et al., 2015), the network is re-

structured after the counter-terrorism attacks.

(Genkin and Gutfraind, 2011) identifies radical

profiles as well as counter-terrorism strategies using

SNA measures and the position of agents in the

network. Agents in (Moon and Carley, 2007) change

its behaviour and goal according to its position in the

space, since they seek for knowledge and resources

distributed over the geographical space. (Berry, et

al., 2004) models an inter-relation between cliques

and agents’ behaviours, that is these two entities

influence each other at the same time. Finally, the

structure of the network in (Ilachinski, 2012)

depends on roles, operations and the battlefield

context.

2.2.3 Network’s Topology

Regarding the structure of the obtained networks in

analysed papers, it is difficult to stablish a consensus

about the actual topology of TN.

(MacKerrow, 2003) (Berry, et al., 2004)

(Tsvetovat and Carley, 2004) (Genkin and

Gutfraind, 2011) (Li, et al., 2015)obtain

cellular/clique based networks. (Raczynski, 2004)

obtain a tree structure. (Moon and Carley, 2007)

follows a spatial structure. (Keller, et al., 2010) uses

a preferential attachment rule so he obtains a scale-

free network. Finally, (North, et al., 2004)

(Ilachinski, 2012) (O'Neil, 2012) allow their model

to create any kind of topology, depending on

parameters or existing prior data.

2.2.4 Multilayer Networks

As discussed before, qualitative approaches are

interested in the type and motivation of relationship

between individuals, and multilayer structures are

the natural way of taking into account these details.

Furthermore, there are more and more mathematical

formalizations of multilayer networks (De

Domenico, et al., 2013).

Regarding terrorist multilayer networks,

(MacKerrow, 2003) defines kinship, religious,

organizational and friendship layer. (Berry, et al.,

2004) observes the following layers according to the

weight or strength of the link between individuals, in

ascending order by strength: world, mosque,

acquaintance, strong bonds and clique. (Ilachinski,

2012) proposes a more complex layer division, with

3 different layers oriented to a counter-terrorism

strategy: genotype (primitive agent behaviour),

phenotype I (emergent agent behaviour) and

phenotype II (emergent squad and force behaviour).

Moreover, agents in his framework perform actions

in two different spaces: physical and information

one. Finally, (Tsvetovat and Carley, 2004) and

(Moon and Carley, 2007) uses a meta-matrix

approach allowing to take into account agents,

knowledges, resources, locations and tasks.

2.2.5 In/out Data

(Berry, et al., 2004)does not precise any input data,

however a similar paper (Berry, et al., 2003)

discusses about using available data from urban

street gangs as an analogy of terrorist groups.

(MacKerrow, 2003) uses census data, interviews and

GIS data in his model in order to initialize attributes

like ethnography, income, religion, educational

degree, etc. Likewise, the allegiance vector is built

from qualitative data. (North, et al., 2004) generates

a TN from a sample of a given network. (Moon and

Carley, 2007) describes the input data set as the

result of an automatic process of analysis of

unclassified documents like newspaper articles or

intelligence reports using AutoMap text analysis

tool. Regarding spatial information, they hand coded

corresponding latitudes and longitudes. (O'Neil,

2012) extracts information from JJATT database

(doitapps.jjay.cuny.edu/jjatt/) in order to define

agent’s roles. Moreover, since his model mimics real

world networks, the Jemaah Islamiyah network from

JJATT was used as initial network. (Keller, et al.,

2010) models counter-terrorism strategies based on

empirical ones. (Ilachinski, 2012) doesn’t put any

data in his model but the topology and dynamic of

his model have been developed following

operational and ground knowledge. (Genkin and

Gutfraind, 2011) does not use any real data as input

of the model, however the results are partially

validated by empirical data and radical profiles are

based on empirical cases of home-grown self-starter

terrorism. This data comes from some available

datasets as Lexis-Nexis, TRC or START among

others. NetWatch package, presented in (Tsvetovat

and Carley, 2005) and (Tsvetovat and Carley, 2004),

An Agent-based Simulation of Extremist Network Formation through Radical Behavior Diffusion

239

generates synthetic TN based on real network data.

As case of study, they show the generative model of

the September 11th hijackers dataset collected by

(Krebs, 2001). Real data is also used to describe the

terrorist attack strategies, that is sequences of tasks,

resource management, etc.

Table 1 sums up the precedents paragraphs.

3 MODEL DESCRIPTION

The model described below is a GAMA

implementation based on (Berry, et al., 2004).

GAMA is a modelling and simulation platform

(Grignard, et al., 2013) that allows modellers and

experts in numerous domains (ecology, social

sciences, etc.) to build complex models without

strong computer sciences skills, thanks to high-level

modelling language, supporting the integration of

multiple level of agency and realistic environments.

Nevertheless, even if some aspects as social spot

stickiness or social similarity (cf. next paragraphs)

are common to (Berry, et al., 2004), the main goal of

this paper is to show a mechanism of network

formation and not only the recruitment of terrorist

through clique detection. Moreover, this paper

focuses on cultural vector or social transfer and

meeting mechanisms.

3.1 Entities

3.1.1 Social Spots

Social spots represent places where behaviour

diffusion takes place. During the simulation, agents

can attend a social spot. In order to choose a social

spot, a stickiness (Berry, et al., 2004) percentage is

computed. This value depends on agent personality

and social spot ideology. A social spot ideology is

defined from its members’ zeal and cultural vector,

and from its own cultural vector. This last value can

represent the personality of the social spot leader, or

some trait well known about the social spot. For

example, one can have a certain a priori about a

social place independently of the existence of a

leader.

3.1.2 Agents

An agent’s personality is defined by fixed attributes

like age, sex, religion, etc. and a cultural vector

similar to the allegiance vector presented by

(MacKerrow, 2003) that describes cultural traits,

that is the opinion about some specific

organizations, countries, etc. Moreover, an agent has

a value of zeal, which is defined as a measure of

agent’s political violence (Genkin and Gutfraind,

2011). Zeal value allows to make a difference

between an extremist passive agent and an active

one. An agent with a very high zeal value is ready to

becomes a terrorist. This is similar to the

disgruntlement value described by (Berry, et al.,

2004). Zeal evolves every time two agents meet, in

function of their current zeal values. For example, if

both agents are radicals, they mutually increase their

zeal.

“Society” is a special kind of agent based on

(Berry, et al., 2004) that influences agents at each

simulation step. This influence has an impact in the

agent’s cultural vector depending on its social

pressure sensibility value. Society agent allows us to

model the macro mechanism taking place in the

network formation process.

3.2 Dynamics

The dynamic of the model is fully described by 3

processes: dynamic related to the social network that

is the formation of links, the attending to a social

spot and the social transfer between agents. These

dynamics are computed sequentially (discrete event

simulation) Every dynamic influences the other,

since for example an agent will create a link with

other agent depending on a score taking into account

if they are in the same social spot and the similarity

between their cultural vectors.

3.2.1 Link Creation

The model allows agents to meet in different ways.

First, they can do random meetings with a certain

probability (Tsvetovat and Carley, 2005). Otherwise,

there are two meeting or dynamic mechanisms:

a. According to the social similarity

Here, a social similarity measure is used to

decide whether two random agents are able to create

(or remove if below a threshold) a link between

them within the social network. Social similarity is

based on the similarity equation presented in (Berry,

et al., 2004) and network measures like shared

friends, homophily and degrees.

b. According to agent’s opinion

The model presented in (Holme and Newman,

2006) is used to compute the dynamics of opinions.

At each simulation step, an agent has two

possibilities: Either to switch one of his links by

another one (that is remove one link and create

another one) that allow it to connect with an agent

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

240

having the same opinion or to replace his own

opinion by the opinion of one of its neighbours,

keeping the same link.

3.2.2 Attending a Social Spot

An agent can belong to a social spot if a “stickiness”

score is high enough to attend it. This score is based

on (Berry, et al., 2004) with a modification allowing

to take into account both fixed characteristic and

cultural vectors.

3.2.3 Social Transfer

The social transfer mechanisms allow agents to

communicate and exchange their cultural values,

what will have an impact in the other dynamics

(creating links and attending a social spot).

The current version of the model described here

uses Axelrod model for the dissemination of culture

(Axelrod, 1997), which allows to change one value

of the cultural vector at each interaction.

4 RESULTS

The following results show how a network is formed

through a radicalization process depending on the

social meeting mechanism used. The red dots

represent radical people and the white ones are

neutral people. For theses simulations, we have

initialized a population of 20 agents, 17 of which are

neutrals and 3 radicals. This size corresponds to a

neighbourhood scale simulation, where opinions can

diverge, be homogenized, evolve, etc. We think this

is a good trade-off between a large-scale simulation

as a region or inter-countries simulation

(MacKerrow, 2003) or a more restricted simulation

within a small group of 5-10 participants as many

sociological works.

Cultural and fixed vectors and zeal values are

initialized randomly between [-1,1], and social

pressure sensibility is minimal in the case of radical

agents, that is they are totally convinced in their

cause and can’t be neutralized. In this simulation,

there is a radical social spot and a neutral one, so its

cultural vectors correspond to its ideologies.

Moreover, the society agent influences some random

agents at each simulation step. This influence

changes some values of cultural vector in order to

radicalize the agent.

As explained above, there are two mechanisms

for agents to meet. The first one, based on a social

similarity metric, generates a network with two main

components: a clique of neutral people and another

one of radicals Figure 1.

This is an expected result since the model is

based on Berry’s (Berry, et al., 2004) which aims to

show the clique based structure of a TN. More

interesting is the evolution of the structure through

the simulation. One can observe that on t=75 there

are two communities of neutral people. This is due

to the existence of two different social spots, but

there is a homogenization process over the

simulation that bring together all neutral people in

one single clique. Concerning radical people, there is

a one person who was radicalized between t=0 and

t=75. Nevertheless, radicals don’t create a network

as quickly as neutrals. This can be explained with

the zeal parameter. Actually, one can be radical

because of different raisons, but the probability of

creating a link between two radicals is higher when

the zeal level is similar, that is when they are ready

to become an actual terrorist, meaning ready to

perform a terror act. There are also some contacts

between radical and non-radical people, but they

become increasingly rare.

Figure 2 corresponds to the network formation

process using Axelrod combined with (Holme and

Newman, 2006). The obtained structure is

completely different compared to the previous one.

People are more dynamic and choose their contacts

giving more importance to the opinion. Opinions

change in a more dynamic way, and the result is a

population with more radicals. Some people keep

their opinion and remain isolated from the main

component, that was not possible with the previous

model. This can be explained because social spots

have less impact on people behaviour, since people

are less devoted to the behaviour supported by social

spots and more on people’s one.

5 CONCLUSIONS

This paper describes a model of terrorism network

construction, with mechanisms allowing the

diffusion of cultural values, social meetings and

radicalization. We showed how the diffusion of

radical behaviour is related to the formation of the

network and vice-versa.

Axelrod model is used for the cultural transfer

between agents. However, other models of social

transfer as (Schleussner, et al., n.d.) (MacKerrow,

2003) (Deffuant, et al., 2002) are being developed

and adapted to our model in order to compare the

resulting networks.

Some important questions in terrorism as

An Agent-based Simulation of Extremist Network Formation through Radical Behavior Diffusion

241

operations, tasks and roles could be added to our

model as separated modules. Similarly, we could

study the multilayer structure of the society,

allowing us to discover more details about the

radicalization process.

Finally, the model is easily adaptable to others

multi-agent platforms or high-level computer

languages

REFERENCES

Axelrod, R., 1997. The dissemination of culture a model

with local convergence and global polarization..

Journal of conflict resolution, 41(2), pp. 203-226.

Berry, N. et al., 2003. Computational social dynamic

modeling of group recruitment, s.l.: Sandia National

Laboratories.

Berry, N. et al., 2004. Emergent clique formation in

terrorist recruitment. s.l., s.n.

Crossley, N., 2010. The social world of the network.

Combining qualitative and quantitative elements in

social network analysis. Sociologica, 4(1), pp. 0-0.

De Domenico, M. et al., 2013. Mathematical formulation

of multilayer networks. Physical Review X, 3(4), p.

041022.

Deffuant, G., Amblard, F., Weisbuch, G. & Faure, T.,

2002. How can extremism prevail? A study based on

the relative agreement interaction model. Journal of

artificial societies and social simulation, 5(4).

Genkin, M. & Gutfraind, A., 2011. How do terrorist cells

self-assemble: Insights from an agent-based model of

radicalization, s.l.: SSRN 1031521.

Grignard, A. et al., 2013. GAMA 1.6: Advancing the art of

complex agent-based modeling and simulation. s.l.,

Springer Berlin Heidelberg, pp. 117-131.

Hidalgo, C. A., 2016. Disconnected, fragmented, or

united? a trans-disciplinary review of network science.

Applied Network Science, 1(1), p. 1.

Holme, P. & Newman, M. E., 2006. Nonequilibrium phase

transition in the coevolution of networks and opinions.

Physical Review, 74(5), p. 056108.

Ilachinski, A., 2012. Modeling insurgent and terrorist

networks as self-organised complex adaptive systems.

International Journal of Parallel, Emergent and

Distributed Systems, 27(1), pp. 45-77.

Keller, J. P., Desouza, K. C. & Lin, Y., 2010. Dismantling

terrorist networks: Evaluating strategic options using

agent-based modeling. Technological Forecasting and

Social Change, 77(7), pp. 1014-1036.

Krebs, V. E., 2001. Mapping networks of terrorist cells.

Connections, 24(3), pp. 43-52.

Lévy, J.-D., Desreumaux, M. & Bartoli, P.-H., 2016.

Baromètre annuel: Bilan de l'année écoulée 2015 et

perspectives pour l'année future 2016, s.l.: Harris

Interactive.

Li, B., Sun, D., Zhu, R. & Li, Z., 2015. Agent Based

Modeling on Organizational Dynamics of Terrorist

Network. Discrete Dynamics in Nature and Society.

MacKerrow, E. P., 2003. Understanding why-dissecting

radical Islamist terrorism with agent-based simulation.

Los Alamos Science, Volume 28, pp. 184-191.

Monge, P. R. & Contractor, N. S., 2003. Theories of

communication networks. s.l.:Oxford University Press.

Moon, I. C. & Carley, K. M., 2007. Modeling and

simulating terrorist networks in social and geospatial

dimensions. IEEE Intelligent Systems, 22(5), pp. 40-

49.

North, M. J., Macal, C. M. & Vos, J. R., 2004. Terrorist

Simulation with NetBreaker. Ann Arbor, Michigan,

s.n.

O'Neil, P., 2012. Dynamic, covert network simulation. s.l.,

Springer Berlin Heidelberg, pp. 239-247.

Raczynski, S., 2004. Simulation of the dynamic

interactions between terror and anti-terror

organizational structures. Journal of Artificial

Societies and Social Simulation, 7(2).

Schleussner, C. F., Donges, J. F., Engemann, D. A. &

Levermann, A., n.d. Clustered marginalization of

minorities during social transitions induced by co-

evolution of behaviour and network structure.

Taylor, M., 2010. Is terrorism a group phenomenon?.

Aggression and Violent Behavior, 15(2), pp. 121-129.

Tsvetovat, M. & Carley, K. M., 2004. Modeling complex

socio-technical systems using multi-agent simulation

methods. KI, 18(2), pp. 23-28.

Tsvetovat, M. & Carley, K. M., 2005. Generation of

realistic social network datasets for testing of analysis

and simulation tools, s.l.: SSRN 2729296.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

242

APPENDIX

Figure 1: Network formation using Axelrod model of social transfer combined with social similarity based mechanism of

meetings.

Figure 2: Network formation using Axelrod model of social transfer combined with Holme’s model (Holme and Newman,

2006) of opinion diffusion.

Table 1: Detail of key features by paper.

An Agent-based Simulation of Extremist Network Formation through Radical Behavior Diffusion

243