A Social Network Model for Integration of Refugees

Fabio Curi

1 a

, Dimitris Nikolopoulos

1 b

and Eric Fernandes de Mello Ara

ujo

1,2 c

Social AI Research Group, Vrije Universiteit Amsterdam, The Netherlands

Department of Computer Science, Universidade Federal de Lavras, Brazil

Keywords:

Social Simulations, Network-oriented Modeling, Refugees Integration, Homophily, Social Network Analysis.

Abstract:

The present work aims to study and model the integration of refugees who appear in a society, through the im-

plementation of adaptive social networks. In our model, for each individual, their characteristics of religious-

ness and language skills are considered for the social integration. We also explore the homophily phenomenon

as part of the creation of new connections and the changes on the refugees’ traits. As it most commonly

happens in real life, refugees appear in a community-structured social network as individual nodes without

any connection, and interactions between refugees and non-refugees are built through a deﬁned methodology

which applies local search, random attachment and node deletion. We show a few case scenarios and perform

a social network analysis.

1 INTRODUCTION

From July to September of 2016, 358.300 ﬁrst time

asylum seekers applied for international protection

in the Member States of the European Union (EU),

mostly Syrians, Afghans and Iraqis. It is undeniable

that the arrival of refugees has impacted the lives of

many countries in the past years, affecting many elec-

tions recently, especially in European countries, such

as The Netherlands, France and recently the US, with

the proposals of new policies to restrain the entrance

of new refugees for the next future (Page, 2016).

As refugees become part of the life of cities

around the world, it brings up an important issue:

their integration in new societies and communities,

especially in regard to the interaction with the locals.

There is a wide range of studies that try to create some

understanding about the processes of settlement of

immigrants ﬂeeing from their home countries (Ager

and Strang, 2008; Korac, 2003; Krahn et al., 2012).

Several approaches have been used in order to analyze

the levels of integration between these two groups,

through case scenarios from speciﬁc countries, dif-

ferent policies regarding immigration and residency

allowance for refugees, etc. (Strang and Ager, 2010;

Fasani et al., 2018).

https://orcid.org/0000-0003-2628-2598

https://orcid.org/0000-0001-6098-6113

https://orcid.org/0000-0003-4263-9075

The discipline of psychology also has much to

contribute to our understanding of immigrants and

the process of immigration. Studies in acculturation

and inter-group relations are focused mainly in two

issues that face immigrants and the society of settle-

ment: maintenance of group characteristics and con-

tact between groups (W. Berry, 2001). Identity strate-

gies employed by immigrants and their counterparts

in the hosting country (especially attitudes toward im-

migrants and toward the resultant cultural diversity)

can result either in a good integration or in marginal-

ization and segregation of the new comers (Sheikh

and Anderson, 2018; Puma et al., 2018).

In the context of analyzing social structures,

network-oriented modeling can be considered as an

alternative way to address complexity for modeling

human and social processes, including the dynamics

of the integration of new people in a pre-established

group (or network) (Treur, 2016). In the present study,

we present an analysis of an adaptive social network

of nodes representing individuals in an environmen-

tal context. The interaction between people is trans-

lated through edges, and dynamic effects such as ho-

mophily are expected to affect the network evolu-

tion over time. As for the application domain, the

aim is to understand, through the model, how the in-

teraction between refugees and their local commu-

nities happens, knowing certain characteristics from

the individuals which might affect these interactions.

The present study applies, as far as known, the ﬁrst

Curi, F., Nikolopoulos, D. and Araújo, E.

A Social Network Model for Integration of Refugees.

DOI: 10.5220/0007930601650175

In Proceedings of the 9th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2019), pages 165-175

ISBN: 978-989-758-381-0

165

methodology in literature directed to studying the in-

tegration of refugees in new communities through

network-oriented modeling.

The following section presents the literature re-

view which supports the decision to use religiosity (or

religiousness) and language domain as the two main

factors for our simulation. The literature also includes

the social network modeling strategies applied in the

simulations. Section 3 presents the temporal-causal

model for the integration of refugees in a society in

more details. Section 4 presents the results of the dif-

ferent simulations. Section 5 presents a technique of

parameter tuning through simulated annealing to bet-

ter ﬁt the model to what we expected to obtain, and

Section 6 includes some concluding comments and

potential future works for improvement in our model.

2 SOCIAL NETWORK

STRUCTURES AND THE

INTEGRATION OF REFUGEES

IN A NEW SOCIETY

In this section, we present the literature explored to

build the model. We explain our choice for language

skills and religiousness as factors that will deﬁne how

connections are built, and also how the social network

is generated based in a realistic approach.

2.1 Factors that Affect Refugees

Integration

The problem of accommodation and integration of

immigrants has been studied from many perspectives

over the years. In the discussion on acculturation

processes, there are three orientations, according to

(Kuhlman, 1991):

[...] those who want all immigrants to adopt

the dominant majority culture, the advocates

of a ’melting-pot’ (i.e. the blending of cultures

and races to produce a new national culture),

and those who favor ethnic pluralism in which

communities retain much of their original cul-

ture and the country becomes a federation of

nationalities.

(Kuhlman, 1991) studied Eritrean refugees in the re-

gion of Kassala (Sudan) and created a framework

to explain the issue of involuntary immigration in

developing countries from an economic perspective.

Their proposed framework has four independent vari-

ables with characteristics of the refugees themselves,

factors related to the process of ﬂight, character-

istics of the region of settlement, and policies re-

lated to refugees. The dependent variable is the in-

tegration. From the characteristics of the refugee

are included (1) demographic variables, (2) socio-

economical background and (3) ethno-cultural afﬁl-

iation.

From the demographic characteristics of refugees,

are listed criteria as age, sex and household compo-

sition. From socio-economical background are in-

cluded educational level, occupation before immi-

gration and a distinction between rural and urban

refugees. For ethno-cultural afﬁliation of refugees,

are included factors as as native tongue, religion and

place of birth.

(Krahn et al., 2012) shows the impact on labour

market for refugees from Yugoslav and Russia in the

90s, and points out that the level of English language

training they received was inadequate to meet the re-

quirements of their occupations, when comparing im-

migrants with Canadian-born workers with the same

level of education and professional degree. In the

work done by (Ager and Strang, 2008), elements cen-

tral to perceptions of what constitutes ‘successful’ in-

tegration in the UK are identiﬁed. The framework

contains four main domains of integration, (1) foun-

dation (right and citizenship), (2) facilitators (lan-

guage, cultural knowledge, safety and stability), (3)

social connection (social bridges, bonds and links)

and (4) markers and means (employment, housing,

education and health). Regarding the social connec-

tions, (Ager and Strang, 2008) shows that both immi-

grants and people within the community had expecta-

tions of a community where there was active ‘mixing’

of people from different groups, with ‘belonging’ be-

ing one of the marks that would identify a successful

integration.

Apart from the view of the need for integration, it

is shown in many studies that the maintenance of tra-

dition brings many beneﬁts to the refugees, like health

and the integration itself (Beiser et al., 1993). The fre-

quency and quality of contact between refugees and

non-refugees is also an important aspect that helps in-

tegration. (Ager and Strang, 2008) states that:

In the course of our ﬁeldwork, both refugees

and non-refugees suggested that an important

factor in making them feel ’at home’ in an

area was the friendliness of the people they

encountered on a daily basis. [...] Con-

versely, perceived unfriendliness undermined

other successful aspects of integration.

The language is also considered as central for the inte-

gration process by (Ager and Strang, 2008), affecting

the social interaction, economic integration and full

SIMULTECH 2019 - 9th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

166

participation in society. Researchers were able to con-

duct longitudinal analysis and learned that language

skills are needed for meaningful contact rather than

vice versa (Collyer et al., 2017). Along with the lan-

guage, refugees’ knowledge of national and local pro-

cedures, customs and facilities and, though to a lesser

extent, non-refugees’ knowledge of the circumstances

and culture of refugees are also relevant.

Religiousness is here expressed by how religious

an individual is, regardless of the religion branch. A

correlation between depression and religious practice

in Lebanon’s elderly is found by (Chaaya et al., 2007).

The sample taken from a refugee camp that had no ac-

cess to the mosque showed a higher level of depres-

sion, due to lack of socialization. It is also shown that

minority groups rely on religious stratagems to cope

with their distress more than other groups do (Dunn

and Horgas, 2004).

As discussed above, in many works within psy-

chology and sociology of refugees, the complexity

of the phenomenon of integration and/or accultura-

tion is big. Tackling all the aspects demands a com-

plexity that does not guarantee a total accuracy in re-

spect to reality. Nevertheless, some factors are present

in many works and can be addressed for a simpli-

ﬁed, yet also relevant perspective over the problem.

For this work, we chose religiousness and language

skills as the factors that will deﬁne integration and

construction of new bonds between refugees and non-

refugees.

2.2 Building a Social Network of

Community and Refugees

Network-oriented modeling is an approach that com-

bines social network analysis in the search of under-

standing phenomena that can be represented as net-

works (Treur, 2016). These applications are seen in

many ﬁelds of study, but recently much is found in

the study of social relations. As one can use nodes

and edges to represent objects and relations, this ap-

proach is appropriately suitable for our objectives.

(A. Beaman, 2012) uses social network analysis to

evaluate how the construction of a social network in-

ﬂuences the success of refugees in the US labor mar-

ket. The results indicate that an increase in the num-

ber of social network members resettled either in the

same year or one year prior to a new arrival leads to a

deterioration of outcomes, while a greater number of

tenured network members improves the probability of

employment and raises the hourly wage.

Both studies by (Beiser et al., 1993) and (Beiser

and Hou, 2001) demonstrate how language skills,

combined with employment, can be used for predic-

tion of depression in Asian refugees. (Chaaya et al.,

2007) also shows how religiosity can be inﬂuenc-

ing depression in older people in Lebanese refugee

camps.

Moreover, many works aim to gather longitudinal

and ego-centered data about refugees and build so-

cial networks for analysis (A. Beaman, 2012; Koser,

1997; Ryan et al., 2008; Williams, 2006). However,

no work addressing simulations and predictions of fu-

ture situations of refugees regarding their integration

using the traits of people in a social network were

found.

So as to mimic a real network with the properties

chosen for the model, it is necessary to build a social

network which can handle weighted edges, and where

homophily can be incorporated to handle the changes

in the node states. The work done by (Toivonen et al.,

2007) is a very suitable and good inspiration for our

work. They present a model which simulates real

networks taking into account the theory of weight-

topology correlations as their basis. The model will

be better explained in the next Section.

3 CREATING A COMMUNITY

AND MANAGING THE TIES

This section explains how the network which repre-

sents the population of a society is created. In order

to do so, we explain the model created by (Toivonen

et al., 2007), as well as the adaptations that were made

in order to adequate the model to our purposes.

3.1 A Model for the Creation of a

Community of People

Social networks are a very suitable approach to repre-

sent a community, where the nodes represent the in-

dividuals of a society and the connections represent

the bonding between these individuals. As human re-

lations change over time, some phenomena might be

implemented, such as nodes dying or leaving the com-

munity, connections being strengthened or weakened,

new nodes becoming part of the network, etc.

The algorithm proposed by (Toivonen et al., 2007)

works over a ﬁxed size of network deﬁned as N nodes,

initially not connected with each other and with no

state values. Afterwards, the algorithm operates over

them by creating and modifying their relations based

on random attachment, local search and node dele-

tion, providing a stabilized network as an output. We

consider a network to be stabilized when it follows the

community structure of a social network with strong

A Social Network Model for Integration of Refugees

167

Figure 1: A social network with 300 nodes and a community structure (left). Each community is represented by a different

color. Next, there is an insertion of 20 isolated nodes without any connections (right), representing the refugees (in gray).

connections inside the communities and weaker ones

connecting the communities (S. Granovetter, 1983).

The local search mechanism is a two-step proce-

dure in which every node has the chance to choose a

neighbor and increment the connection between them

(ﬁrst step). Then, it proceeds by choosing one of the

neighbors of the ﬁrst chosen node, enabling their con-

nection to be strengthened (second step). In the sec-

ond step, if the node and the neighbor of its chosen

neighbor are already connected, then their connection

is strengthened by an increment value. If they are not

connected, then they have a chance to connect with

each other. This corresponds to the real-life situation

in which if a person has strong ties with two others,

then these latter two have a high chance to get con-

nected. This is also known as the weak tie hypothesis,

introduced by (Rapoport, 1957).

The random connections that we can create in real

life are represented through the random attachment

mechanism. In the algorithm proposed by (Toivonen

et al., 2007), if a node is not connected with any other

node after one step of ﬁnding new friends at random,

then it gets connected to a random one.

The last procedure is the node deletion, through

which a node has a slight chance to be deleted in ev-

ery step. If a node gets deleted, then its connections

are deleted as well. The deletion of nodes can be in-

terpreted as a mechanism of preventing the network

from becoming a clique, whilst it also represents, for

the dynamic network, link removals between nodes,

thus a sudden breakdown of the relationship between

two nodes, as in real life relationships can end. In

order to prevent the elimination of the network, if a

node is deleted then a new one appears in the next

time step.

Some described parameters were considered nec-

essary and are listed as follows (Toivonen et al.,

2007):

: probability of a node to establish a new link with

another randomly chosen node (used in random

attachment);

∆

: probability of two nodes connected to the same

node, to be connected (used in local search);

: probability of a node to be deleted (used in node

deletion);

δ : weight increment;

: initial weight of a new connection.

For our simulations, we kept the original prob-

ability values provided by (Toivonen et al., 2007),

= 0.0005, p

∆

= 0.05 and p

= 0.001. The val-

ues for the weight of the edges have been normalized

between the range of [0,1]. Therefore, we have de-

cided to keep low values for δ and ω

, with values of

0.0001 and 0.0001, respectively. These value choices

are small enough so that the weights of the edges do

not exceed 1 rapidly throughout the simulation life-

time. Nonetheless, these values can be adjusted in

order to represent reality as accurately as possible. If

people tend to increase their bonding faster, then δ

should be higher, otherwise it should be lower. Also,

can variate depending on the initial bonding level

connecting people. For the purposes of the current

case of study, it was decided to keep these values

static as 10

−4

The algorithm runs these three procedures for N

nodes and for a speciﬁc number of steps. Figure 1

on the left illustrates the output graph for a simula-

tion after running it for 25.000 steps upon N = 300

SIMULTECH 2019 - 9th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

168

Table 1: Traits of the populations inside the community.

Random values are uniformly distributed within the ranges.

Community Lang. skills Religiousness

Refugees [0.0,0.2] [0.7,1.0]

Local population [0.6,1.0] [0.0,0.4]

nodes. The network has a community structure and

each community is shown in a different color. The

total number of communities in this example is 21.

Results showed that strong connections are occurring

between nodes in the same communities and weaker

links connect the communities. On the right side,

it shows the same social network of 300 nodes af-

ter the insertion of 20 isolated nodes representing the

new refugees recently arrived in this particular soci-

ety. From this status, we included our mechanisms to

cope with the homophily effects in the relationships

within the community.

3.2 Incorporating Traits of Language

Skills and Religiousness

After the initial community is established, the lan-

guage skills and religiousness values in the nodes are

set, representing local population and refugees ac-

cording to Table 1, at random. As expected, the lan-

guage skills of the local population is higher than the

recently arrived refugees. Moreover, the religiousness

of each node from the local population is also much

smaller than the refugees population, simulating the

context of a secularized society receiving immigrants

from religious areas.

After attributing values for these two traits to all

the nodes in the population, we incorporate the effects

of homophily and social contagion in each node over

time. The mechanisms of random attachment, local

search and node deletion are still running, but now the

state of the nodes is affected by their relationships.

3.3 Numerical Representation of the

Homophily Principle for the Weight

of the Edges

The values for the weights of the edges are updated

over time using homophily, which indicates that the

more similar the states of two connected nodes are,

the stronger their connection will become (Treur,

2016). Node values alone are not representative of

the homophily effect, but their absolute differences

are. The values of two connected nodes can be here

generalized as X

and X

. Furthermore, a thresh-

old value τ

A,B

is understood as being the reference

value above/under which the absolute differences of

node values should be so that upward or downward

weight changes occur, respectively. These thresh-

olds are present in Equation 3 when calculating the

changes in the connection weights and have a ﬁxed

value of 0.1. Their role is explained as follows:

• an upward change of the connection weight ω

A,B

occurs when |X

− X

| < τ

A,B

• no change of the connection weight ω

A,B

occurs

when |X

− X

| = τ

A,B

• a downward change of the connection weight

A,B

occurs when |X

− X

| > τ

A,B

The update of the the edge weights is calculated

through Equation 1, given the actual weight of the

connection between two nodes A and B at a time-step

t, ω

A,B

(t), a speed-factor η

A,B

, and a function c

A,B

rep-

resenting the combination of the aggregated impact

caused by B in A.

A,B

(t +∆t) = ω

A,B

(t) + η

A,B

(t), X

(t), ω

A,B

)

−ω

A,B

(t)]∆t

(1)

The advanced quadratic function based on (Sharpan-

skykh and Treur, 2014) was chosen to update the edge

weights, as shown in Equations 2 and 3. X

and X

are the states of person A and person B, respectively,

D = |X

(t) − X

(t)| is the difference between the two

node states, η

A,B

is the update speed parameter for the

connection from person A to person B, and τ

A,B

is the

threshold for connection adaptation.

A,B

,ω

A,B

) = h

A,B

(D,ω

A,B

) (2)

A,B

(D,W ) = W + Pos[η

A,B

(τ

A,B

− D

)](1 −W )

−Pos[−η

A,B

(τ

A,B

− D

)]W

(3)

In Equation 3, Pos(x) = (|x| + x)/2, which returns x

when x > 0 and 0 otherwise.

3.4 Numerical Representation of the

Social Contagion for the State of the

Nodes

In parallel to the weight evolution, the node values for

the states are also changing. The effects of social con-

tagion are expressed in the evolution of node values.

The modeling of such phenomenon is given through

the differential Equation 4:

(t +∆t) =

(t) + η

(t), .. ., X

(t)) − X

(t)]∆t

(4)

A Social Network Model for Integration of Refugees

169

, where X

is a characteristic value for node B, η

is the speed factor which indicates how fast this node

can change its value, and c

is a combination function

which takes into account the inﬂuence of all nodes

A1,...,Ak connected to B, with k being the degree of

node B, as shown in Equation 5.

For our simulations, we used the advanced nor-

malized sum combinational function c

, adnorsum,

shown in Equation 6.

(ω

A1,B

(t), .. ., ω

Ak,B

(t)) =

adnorsum(ω

A1,B

(t), .. ., ω

Ak,B

(t))

(5)

adnorsum(ω

A1,B

(t), .. ., ω

Ak,B

(t)) =

∑

i=1

∗ ω

Ai,B

∑

i=1

Ai,B

(6)

As before-mentioned, two characteristics are con-

sidered and expected to inﬂuence the interaction be-

tween the individuals in the network: language skills

and religiousness. Other aspects could be possibly

included in order to make the model more realis-

tic, but for the sake of simplicity we only considered

these two states for evaluation. We also believe that

the inclusion of new states would possibly enrich the

model and could be an interesting exploration for fu-

ture work.

Next, we have studied different scenarios combin-

ing both aspects in the integration of the refugees.

4 MODEL SIMULATIONS AND

DISCUSSION

This section presents the results for the simulations

performed considering the model presented. As we

have two traits affecting the nodes through social con-

tagion and the homophily effect changing the weights

of the edges, we divided the simulations in order to

better comprehend the outcomes. Important parame-

ters were initialized with certain values in order to ob-

tain results as close to reality as possible. These last

are: the number of nodes, the time window frame,

the step size and the speed factors. The number of

nodes was kept low, having 300 non-refugees and a

certain number of refugees appearing in the popula-

tion. This last number can, of course, be adjusted to

simulate different case scenarios by, for instance, get-

ting the percentage of refugees in a speciﬁc country.

As the proposed model here is a general one, we have

not set values related to any speciﬁc society. The time

frame has a value of 36, and the step size is 1. These

values are representing a time step of a month, and it

is assumed that within 36 months each person inter-

acts with others at least each month. The speed fac-

tors for both the nodes and the edges update is kept at

0.1 since it is considered that religiousness, language

and relationships change slowly over time. Here, one

must bare in mind that relations are also affected by

the local search method which runs without any speed

factor. The speed can actually be expressed inside the

increase rate δ: the lower it is, the slower the relation-

ships are being incremented.

During the lifetime of the simulation, some ran-

dom nodes of the network have the chance to be

deleted from it and ’move away’ from our observ-

able world. Despite their disappearance, these nodes

are represented in the output graphs, because they

play their role in the evolution of the overall relation-

ships and node state values, regardless of their signif-

icance. It can be observed that, in the output graphs

of the node state and weight values throughout time,

some lines are abruptly cut at speciﬁc points. That

means that a node has been removed from the net-

work, alongside with its links. This can be corre-

lated to real-life phenomena, such as a person mov-

ing away from a society or even unfortunately pass-

ing away. Similarly, some lines might appear at time

points which are different from zero. Of course, this

is an expected behavior because each time a node is

deleted, another one is added in the following step.

The addition of these nodes prevents the shrinking of

the network, which would be inevitable if the nodes

kept being deleted throughout a large number of steps.

4.1 Scenario 1 - Language Proﬁciency

and Homophily

In this scenario, we excluded the effect of religious-

ness as we are interested in knowing what happens

with the population of refugees over time concerning

their language skills, and how is the integration based

only on this trait.

The output plots for the language skill evolution

over time are shown in Figure 3 (left). One can see

that most of the refugees are improving their language

skills and are fully integrated in society. Nonetheless,

if the number of refugees is increased, then it is ex-

pected that some of them will not be able to be inte-

grated in the society: at least not as quickly. In fact,

it can be observed that many of them remain isolated

without the ability to get integrated.

Figure 3 (right) shows the evolution of the weights

throughout time. As a reminder note, the lines that

were abruptly cut represent the node deletions. So as

not to visually damage the plots, it was decided not

to show whenever these values drop to 0 from one

SIMULTECH 2019 - 9th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

170

Figure 2: The ﬁnal graph produced for language skills after

running the simulation.

simulation step to the other. Note that similar results

from the ones obtained in Figure 3 are valid for this

algorithm for whichever size of population.

Figure 2 shows the resulting graph of this network.

Some refugees are integrated in the social network,

although others are isolated, such as nodes r6, r12 and

r21 (in dark green).

4.2 Scenario 2 - Religiousness and

Homophily

Now, the effect of language skills is excluded as we

are interested in discovering what happens with the

population of refugees over time concerning their reli-

giousness, and how is the integration done based only

on this trait.

The node and edge updates are shown in Figure

4. The results are analogous to the ones of languages

skills, except for the fact that now the values of reli-

giousness converge towards the most common values.

Nonetheless, some refugees remain isolated and still

hold a quite stable value for their religiousness. A

speed factor of 0.05 was used for the homophily ef-

fect.

As it can be seen, the node values for the refugees

have dropped considerably within the simulation

time, whilst the local community remained with val-

ues rather unchanged, or very little. The religiousness

of the former group tends to go down after the social

network is run as expected. However, it is also ex-

pected that this parameter does not ﬂuctuate as much

as language skills, for instance. The refugees’ level of

religiousness tends to go down after their integration

in the society and thus, interacting with the local com-

Table 2: Degree distributions of the refugee nodes.

Degree 1 2 3 4 5

Refugees (%) 10 20 40 20 10

munity, which in turn has smaller religiousness val-

ues. Furthermore, the religiousness values of the lo-

cal community had very little variation in time, which

is explained by the combination of the low speed fac-

tor for the node values update and the less inﬂuential

presence of the refugees if compared to the total num-

ber of locals in the society. Note that similar results

from the ones obtained in Figure 4 are valid for this

algorithm for whichever size of population.

In Figure 5, the ﬁnal graph of the second scenario

is shown. It is thus now harder to identify the loca-

tion of these refugees, which conﬁrms that most them

managed to integrate in the network.

4.3 Social Network Analysis

The aim of the present section is to obtain rele-

vant statistics of the proposed network. We built a

model of 300 local individuals and 10 refugees be-

ing inserted in the society. These individuals are

distinguished by their labels ‘p’ and ‘r’, respec-

tively. Religiousness is speciﬁcally taken into ac-

count, where high values are observed virtually only

by the refugees, even after running of the algorithm.

However, certain local individuals were observed to

have increased their religiousness values through the

interactions in the social network. High values are

mainly observed around the refugee nodes, which

explains that the homophily effects were signiﬁcant

enough to alter node values. In Table 2, the degree

values of the nodes are depicted against the number

of occurrences. The degree values are representative

of how connected they are to the network.

The overall average degree was observed with

value 10.26. As expected, the curve follows a rather

downward exponential distribution, with a few out-

liers present around the theoretical curves (i.e. the

occurrences with values 1, 2 and 3). This result is ap-

proximately close to a power law distribution, which

indicates that our network is close to being scale-free.

Table 2 shows the results for the refugee population.

The next step is to obtain the centrality of nodes in

the network, thus, the relative importance of the nodes

is obtained. In the model, we deﬁned that a stabilized

model is one that follows certain properties, such as

being scale-free and having the shortest path-length

with rather low values. The average path length had

a value of 3.75. The eccentricity distribution was also

calculated, which measures the distance from a given

A Social Network Model for Integration of Refugees

171

Figure 3: Evolution of language skills (left) and connection weights (right) in a population of 350 nodes, among which 50 are

refugees using adnormsum for (left) and advanced quadratic functions for (right).

Figure 4: The evolution of religiousness for each node using adnormsum (left) and for the connection weights for each pair

of connected nodes using advanced quadratic as combination function (left).

starting node to the farthest node from it in the net-

work. The biggest proportion of the individuals in the

network have eccentricity distribution values around

8, which indicates that this same number of individ-

uals are needed to connect the respective node to an-

other extreme of the network. However, a few occur-

rences needed a much larger number of connections

to reach these same extremities. Table 3 depicts the

eccentricity values for the refugee population.

In Figure 6, eccentricity values are shown within

the network. The nodes in dark gray and pink are

those with highest eccentricity values. Surprisingly,

these individuals are not refugees, and we obtain here

interesting results which show that these last were cat-

egorized in another set of values, mostly 7,8 and 9, as

shown in Table 3. Even though these eccentricity val-

ues are somewhat high, they are representative of the

success of our algorithm to integrate these individuals

in the society.

Regarding the different communities, as the social

network is formed, the initial refugee community is

expected to spread as the connections with local indi-

viduals develop. A total of 14 communities were ob-

served and the smallest and biggest ones had 2 and 56

members, respectively. One interesting observation is

that the refugees were now spread around in the net-

work: in certain communities, there were occurrences

of a single refugee present. In others, not many coex-

isted. The task of integration has been successfully

carried out by implementing both algorithms at the

same time. Finally, so as not to render this section

exhaustive, only religiousness was considered, given

that similar trends for language skills are expected,

except from the fact that these last are expected to

increase as the homophily effects take effect in the

algorithm. Furthermore, as the speed factor for lan-

guage skills is considerably higher, higher degree dis-

tributions and lower eccentricity values are expected

if compared to religiousness.

5 PARAMETER TUNING

The second part of the overall analysis of the model

behavior is focused on validating the proposed model

through parameter tuning. Due to the fact that any

model must be a close approximation to the real-life

phenomenon that it corresponds to, it must be “tuned”

in such a way that it is as representative and reliable as

possible. This is done through choosing speciﬁc pa-

rameters and applying optimization techniques upon

them in order to obtain the best corresponding values.

For that reason, these techniques are based on

some empirical data that is given as an input and is

compared to the observed data that the model outputs.

The main goal is to obtain as little erroneous behavior

as possible, with the error being deﬁned as the mean

squared error between the simulated and the empirical

SIMULTECH 2019 - 9th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

172

Figure 5: The ﬁnal graph produced for religiousness after

running the simulation.

Figure 6: Eccentricity of religiousness in the social net-

work.

Table 3: Eccentricity distributions of the refugee nodes.

Eccentricity 7 8 9

Refugees (%) 20 60 20

values.

Simulated annealing (Eglese, 1990) is used to

ﬁne-tune the model and make it more representative

and reliable. Pseudo-empirical values of the state and

weight values of the population were created, as the

extraction of reliable related real data was hard. As

it is known, simulated annealing combines the advan-

tages of gradient descent and random walk in such

a way that it assures efﬁciency and completeness.

Three parameters of it are crucial for its outcome on

parameter tuning optimality: the temperature T , the

minimum temperature T

min

, the parameter α, the max-

imum number of iterations i, and the deﬁnition of the

acceptance probability. The temperature T is normal-

ized between the values 0 and 1, and consequently the

starting value T is set to 1. At each step, T is multi-

plied by α, shrinking the value and getting closer to

min

, which is set to 0.001. We set value α to be 0.9,

which guarantees to give a satisfactory number of it-

erations which would lead to optimal results, however

sacriﬁcing run time. The number of iterations at each

temperature step is set to be 100. Finally, the accep-

tance probability method is given based on the val-

ues of the temperature, the new cost and the old cost,

where new cost > old cost. Even if the new cost is

higher than the older cost, it is accepted with the fol-

lowing acceptance probability:

acceptance

= e

−

newcost−oldcost

(7)

All the above parameter values are believed to give

the best expected results, or at least results with a

neglibilble arbitrary error. The pseudo-data was ex-

tracted by adjusting the original model parameters in

such a way that the output seems as realistic as pos-

sible. For instance, we adjusted the speed factor and

the rest of the functional parameters in a way that the

religiousness of refugees would slowly decline over

time. As the time frame is set to show three years,

it is believed that the religiousness of refugees would

not be totally adjusted to the general population, since

it is a characteristic that is deeply rooted and it is

not expected to change signiﬁcantly. The extracted

pseudo-empirical data consists of ’.csv’ ﬁles which

contain the values of the states and the edges through-

out time. In order to simplify the procedure of sim-

ulated annealing in terms of run time, these values

were extracted in each 6 time steps, starting from time

points 0 and 5 and moving on to points 11, 17, 23,

29 and 35. We chose a small number of parameters

in order to obtain results within reasonable time, and

excluded the error calculation between edges values,

as this would make the program run within unreason-

able time. Nevertheless, it can be easily extended by

adding more parameters and adjusting the error cal-

culation.

The initial parameters chosen were the speed fac-

tor and the threshold used within the functions that

apply the homophily principle. The speed factor is

used by both combinational functions in the edge up-

date and the node update, respectively, and is set to

an initial value of 0.5. The threshold value is used by

the edge update’s advanced quadratic function and is

also initially set to 0.5. After running the simulated

annealing for a model based on the language level

A Social Network Model for Integration of Refugees

173

of 2 refugees among a society of 10 non-refugees,

we obtained satisfactory results. The ﬁnal run of the

model under simulated annealing gives a diagram for

the evolution of the node states which is depicted in

Figure 7. The x-axis gives the time frame measured

in months. In this particular instance, one can see that

the refugees are fully integrated after the 25th month.

This model is really close to the one obtained by the

empirical data.

Figure 7: Evolution of language skills after parameter tun-

ing through simulated annealing.

6 CONCLUSIONS

The present work aimed at studying the integration

of refugees through the implementation of a network-

oriented model. Individuals from local communi-

ties and refugees were represented as nodes and rela-

tionships between them were modeled as edges with

weight values varying in time. The signiﬁcance of

each individual was made through values represent-

ing religiousness and language skills. As it is a model

based in real-life scenarios, these values were created

as close to reality as possible, indicating low levels of

language skills for refugees arriving in the commu-

nities, for instance. Notwithstanding, minimum and

maximum values were set, being a constraint for the

model.

A model proposed in literature was used so as

to build a stabilized network, upon which a dy-

namic model was implemented in order to render

node and edge values dynamic. Two case scenarios

were explored, considering two characteristics from

the individuals: religiousness and language skills.

A set of combination functions were selected so as

to implement the evolution of node and edge values

within time. Social contagion and homophily effects

were best represented by the adnorsum and advanced

quadratic functions, respectively, as they were closer

to a real-life scenario. Furthermore, node deletion

and node creation were both implemented as they are

characteristics of a real society: relationships end and

start continuously in a population setting. The param-

eters of the dynamic model were adapted to represent

how these interactions evolve in time. It is expected

that people interact with each other at least once every

3 years, and speed factors were rather low, as it is ex-

pected that language skills and religiousness change

slowly with time. Different scenarios were created,

in which a certain number of refugees arrive in these

local communities. It was observed that integration

was observed for most of them, however a part of the

refugee population did not manage to become inte-

grated in the society, and levels of language skills and

religiousness stayed rather unchanged in time.

A social network analysis was also built, which

studied more in-depth the evolution regarding the

refugee population. Degree and eccentricity distribu-

tion measurements were calculated in order to better

understand how integrated these individuals became.

Later on, simulated annealing was successfully ap-

plied as a parameter tuning technique based on em-

pirical data. The implemented code worked upon the

state values and tried to tune both speed factor and

threshold, and can be safely extended in the future in

order to run upon more parameters and error calcula-

tions.

Finally, the present work implemented a method-

ology of manipulating stabilized networks in order to

build case scenarios in which new individuals appear

in a much larger setting of people. The interest being

to understand how these interactions are built and de-

veloped throghout time, the proposed methodology is

an efﬁcient and feasible way of studying the integra-

tion of refugees in their new communities.

Other interesting scenarios would be to study how

the extroversion and openness of both the local popu-

lation and the refugees would help these last integrate

in a society. As explained by (Dinesen et al., 2014),

openness has an unconditional effect on attitudes to-

ward immigration: scoring higher on this trait implies

a greater willingness to admit immigrants, thus inter-

acting with them once they have already reached their

ﬁnal refuge destinations. Expressiveness would be

understood as extroversion, which in the model could

be represented as a higher or lower easiness of ampli-

fying one’s social network, by i.e. meeting new peo-

ple and widening contacts within and between com-

munities. Openness and expressiveness values would

be compared alongside, which means that the values

of one characteristic must be compared to the values

of the other one, as it is expected that one’s higher ex-

troversion is best match with another’s higher open-

ness towards others. Thus, in such scenarios, it is ex-

pected that these connections would become stronger

with time.

SIMULTECH 2019 - 9th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

174

REFERENCES

A. Beaman, L. (2012). Social Networks and the Dy-

namics of Labour Market Outcomes: Evidence from

Refugees Resettled in the U.S. Review of Economic

Studies, 79:128–161.

Ager, A. and Strang, A. (2008). Understanding Integration:

A Conceptual Framework. Journal of Refugee Stud-

ies, 21.

Beiser, M. and Hou, F. (2001). Language Acquisition, Un-

employment and Depressive Disorder among South-

east Asian Refugees: A 10-Year Study. Social Science

& Medicine, 53:1321–1334.

Beiser, M., Johnson, P., and Jay Turner, R. (1993). Un-

employment, underemployment and depressive af-

fect among Southeast Asian refugees. Psychological

medicine, 23:731–43.

Chaaya, M., Sibai, A., Fayad, R., and El-Roueiheb, Z.

(2007). Religiosity and depression in older peo-

ple: Evidence from underprivileged refugee and non-

refugee communities in Lebanon. Aging & mental

health, 11:37–44.

Collyer, M., Brown, R., Morrice, L., and Tip, L. (2017).

Putting refugees at the centre of resettlement in the

UK. Forced Migration Review, 54.

Dinesen, P., Klemmensen, R., and Nørgaard, A. (2014).

Attitudes Toward Immigration: The Role of Personal

Predispositions. Political Psychology, 37.

Dunn, K. and Horgas, A. (2004). Religious and nonre-

ligious coping in older adults experiencing chronic

pain. Pain management nursing : ofﬁcial journal of

the American Society of Pain Management Nurses,

5:19–28.

Eglese, R. (1990). Simulated Annealing: A Tool for Op-

erational Research. European Journal of Operational

Research, 46:271–281.

Fasani, F., Frattini, T., and Minale, L. (2018). (the struggle

for) refugee integration into the labour market: Evi-

dence from europe.

Korac, M. (2003). Integration and How We Facilitate it A

Comparative Study of the Settlement Experiences of

Refugees in Italy and the Netherlands. Sociology, 37.

Koser, K. (1997). Social Networks and the Asylum Cycle:

The Case of Iranians in the Netherlands. The Interna-

tional migration review, 31:591–611.

Krahn, H., Derwing, T., Mulder, M., and Wilkinson, L.

(2012). Educated and Underemployed: Refugee

Integration into the Canadian Labour Market.

http://lst-iiep.iiep-unesco.org/cgi-bin/wwwi32.exe/

[in=epidoc1.in]/?t2000=025393/(100), 1.

Kuhlman, T. (1991). The Economic Integration of Refugees

in Developing Countries: A Research Model. Journal

of Refugee Studies, 4.

Page, E. (2016). European Commission Statis-

tics Agency Eurostat (2016) Asylum quarterly

report. http://ec.europa.eu/eurostat/statistics-

explained/index.php/Asylum quarterly report.

Accessed: 2017-05-22.

Puma, J. E., Lichtenstein, G., and Stein, P. (2018). The rise

survey: Developing and implementing a valid and reli-

able quantitative measure of refugee integration in the

united states. Journal of Refugee Studies, 31(4):605–

625.

Rapoport, A. (1957). A Contribution to the Theory of Ran-

dom and Biased Nets. The Bulletin of Mathematical

Biophysics, 19:257–277.

Ryan, L., Sales, R., Tilki, M., and Siara, B. (2008). So-

cial Networks, Social Support and Social Capital: The

Experiences of Recent Polish Migrants in London.

Sociology-the Journal of The British Sociological As-

sociation - SOCIOLOGY, 42:672–690.

S. Granovetter, M. (1983). The Strength of Weak Ties:

A Network Theory Revisited. Sociological theory,

1:201–233.

Sharpanskykh, A. and Treur, J. (2014). Modelling and

Analysis of Social Contagion in Dynamic Networks.

Neurocomputing, 146.

Sheikh, M. and Anderson, J. R. (2018). Acculturation pat-

terns and education of refugees and asylum seekers: A

systematic literature review. Learning and Individual

Differences, 67:22–32.

Strang, A. and Ager, A. (2010). Refugee integration:

Emerging trends and remaining agendas. Journal of

refugee studies, 23(4):589–607.

Toivonen, R., M. Kumpula, J., Saram

aki, J., Onnela, J.-P.,

Kert

esz, J., and Kaski, K. (2007). The role of edge

weights in social networks: modelling structure and

dynamics. Proc SPIE, 6601.

Treur, J. (2016). Network-oriented modeling. Springer.

W. Berry, J. (2001). A Psychology of Immigration. Journal

of Social Issues, 57:615 – 631.

Williams, L. (2006). Social Networks of Refugees in the

United Kingdom: Tradition, Tactics and New Com-

munity Spaces. Journal of Ethnic and Migration Stud-

ies - J ETHN MIGR STUD, 32:865–879.

A Social Network Model for Integration of Refugees

175