Conference PaperPDF Available

Modeling Cultural Segregation of the Queer Community Through an Adaptive Social Network Model

Authors:

Abstract and Figures

In this study, the forming of social communities and segregation is examined through a case study on the involvement in the queer community. This is examined using a temporal-causal network model. In this study, several scenarios are proposed to model this segregation and a small questionnaire is setup to collect empirical data to validate the model. Mathematical verification provides insight in the model's expected behaviour. 1 Introduction In a developed world that contains increasingly pluralistic and diverse societies, the establishment of subcultures seems inevitable. In the West, subcultures are associated with an increased identification with in-group individuals, leading to a caring environment. However, subcultures are at risk of a high degree of segregation and misunderstanding of and by out-group individuals, possibly leading to discrimination and aggression. In view of stimulating and maintaining peaceful and democratic processes in these societies, it can be useful to investigate this behavior. This can be performed through Network-Oriented Modelling, which describes the behaviors and opinions of people in relation to the connections among them. In graphical representations of social networks, individuals and their states are depicted by nodes which are connected by uni-or bidirectional links. In this paper, firstly some background about cultural segregation of the queer community is discussed. Next, the network model is explained including the homoph-ily and social contagion principle. In addition, some different scenarios for the model are setup and different simulations with the model are discussed. It will be shown how mathematical verification clarifies how different parameters influence the outcomes of the model. The current paper investigates subculture identification and cultural segregation of the queer community. Being born differently from the heteronor-mative society that they grow up in, queer people have to deal with a number of factors that contribute to a permanent level of distress. Meyer [9] frames this psychological distress as minority stress. Due to a continuous internalized homophobia, stigma,
Content may be subject to copyright.
Modeling Cultural Segregation of the Queer Community
through an Adaptive Social Network Model
Pieke Heijmans, Jip van Stijn, Jan Treur
Behavioural Informatics Group, Vrije Universiteit Amsterdam
piekeheijmans@gmail.com jipvanstijn@gmail.com j.treur@vu.nl
Abstract In this study, the forming of social communities and segregation is
examined through a case study on the involvement in the queer community.
This is examined using a temporal-causal network model. In this study, several
scenarios are proposed to model this segregation and a small questionnaire is
set-up to collect empirical data to validate the model. Mathematical verification
provides insight in the model’s expected behaviour.
Keywords: queer community, temporal-causal network, social hardship, social
contagion, homophily principle, social network, cultural segregation
1 Introduction
In a developed world that contains increasingly pluralistic and diverse societies, the
establishment of subcultures seems inevitable. In the West, subcultures are associated
with an increased identification with in-group individuals, leading to a caring envi-
ronment. However, subcultures are at risk of a high degree of segregation and misun-
derstanding of and by out-group individuals, possibly leading to discrimination and
aggression. In view of stimulating and maintaining peaceful and democratic processes
in these societies, it can be useful to investigate this behavior. This can be performed
through Network-Oriented Modelling, which describes the behaviors and opinions of
people in relation to the connections among them. In graphical representations of
social networks, individuals and their states are depicted by nodes which are connect-
ed by uni- or bidirectional links.
In this paper, firstly some background about cultural segregation of the queer
community is discussed. Next, the network model is explained including the homoph-
ily and social contagion principle. In addition, some different scenarios for the model
are set-up and different simulations with the model are discussed. It will be shown
how mathematical verification clarifies how different parameters influence the out-
comes of the model. The current paper investigates subculture identification and cul-
tural segregation of the queer community. Being born differently from the heteronor-
mative society that they grow up in, queer people have to deal with a number of fac-
tors that contribute to a permanent level of distress. Meyer [9] frames this psychologi-
cal distress as minority stress. Due to a continuous internalized homophobia, stigma,
which relates to society’s expectations, and experiencing actual discrimination or
violence, minority stress is established. She found a strong connection between expe-
riencing minority stress and dealing with psychological distress. She adds that hiding
and concealing, expectation of rejection and ameliorative coping processes contribute
to this psychological minority stress as well [10].
To deal with psychological distress, a social support system can help. For example
successful coming-out stories of other queer people can help reduce the anxiety of
getting negative reactions from friends and family. Wright and Perry claim that sup-
port systems are necessary as they influence the development of young people’s self-
concept and self-esteem [10]. Queer community and queer community spaces can
function as this social support system. Beemyn [2] examines the historical role of
queer spaces and states that queer people needed their own space, not only to escape
from governmental pressures such as police harassment, but also to not deal with
constant territorial struggle, a place where they could escape the dominant cultural
order. These processes lead to cultural segregation, as consequently queers will dis-
tantiate themselves from the dominant cultural order by collectively grouping together
in their own communities. Another strong reason to get involved in such a community
is collectivism, in the sense that people want to benefit their group. The more you are
involved with and identify with this subgroup, the greater your sense of collectivism
[1], and thus the stronger the effect of cultural segregation will be.
From these theories, we expect queer people with a greater experience of hardship
and psychological distress to get involved more in the queer community as the com-
munity serves as a support system, resulting in a stronger cultural segregation. In
contrast, queer people that experience no to little hardship are not inclined to look for
a social support system - however they may be involved with the community for other
reasons, for example relating to their peers. Finally, straight people that experience
hardship may look for community support, but not necessarily for the queer commu-
nity as they do not particularly identify with this group and do not have a strong sense
of collectivism. Resulting from these conclusions, it can be expected that a certain
group of queer people will get involved with their community, finding comfort, sup-
port and finding equals. In contrast, straight people will not share these needs and will
not identify strongly with the queer community. The result is cultural segregation, in
which queer people’s identification with their community is opposed to straight peo-
ple’s identification.
These processes were analysed computationally by designing an adaptive network
model based on the homophily principle that describes bonding between persons that
consider each other similar in some respect(s); see, for example, [8]. This principle
works in combination with the principle of social contagion [3] in a circular mutual
causal relationship, also called co-evolution [5, 18]. In Section 2 the adaptive tem-
poral-causal network model based on these two principles is introduced. Section 3
illustrates the model by example simulations. In Section 4 it is shown that the simula-
tion outcomes are in accordance with what is predicted by a mathematical analysis of
the model. Section 5 describes validation of the model by comparing simulation out-
comes to empirical data and apply Parameter Tuning. Finally, Section 6 is a conclu-
sion.
2 The adaptive temporal-causal network model
A Network-Oriented Modeling approach based on temporal-causal networks [13, 14,
17] was used to analyse the type of processes described in Section 1. This approach
can be considered as a branch in the causal modelling area which has a long tradition
in AI; e.g., see [6, 7, 11]. It distinguishes itself by a dynamic perspective on causal
relations, according to which causal relations exert causal effects over time, and these
causal relations themselves can also change over time. The type of network models
that form the basis is called a temporal-causal network model. These network models
can be used to translate informally described theories from a variety of human-
directed disciplines into adaptive and dynamical numerical models. It takes into ac-
count states and their causal effects on other states. The strengths of causal relations
from a state X to a state Y are indicated by differences in connection weights X,Y.
These connection weights can be combined with activation levels Y(t) of states Y and
used as input for combination functions cY(…) to determine the aggregated impacts on
the states. The precise dynamics of the network are also defined using speed factors
Y of states Y. The network becomes adaptive when connection weights are dynamic
as well. The conceptual representation basically is a graph of states and their causal
relations; a graphical overview of the network is represented as depicted in Fig. 2.
The numerical representation is a translation of this conceptual representation in the
way described in Table 1.
Table 1: From conceptual representation to numerical representation of a temporal-causal
network model; adopted from [17]
Thus, the following difference and differential equation for state Y are obtained:
Y(t + Δt) = Y(t) + ηY [cYX1,YX1(t), …, ωXk,YXk(t)) - Y(t)]Δt (1)
dY(t)/dt = ηY [cYX1,YX1(t), …, ωXk,YXk(t)) - Y(t)]
The adaptive social network model used here is based on two fundamental principles.
Firstly, the notion of social contagion is used to explain the causal influence of one
state on another through the connection between the two [3]. This principle accounts
for the change of state values over time, and its numerical representation is (where XAi
and XB are the states of persons Ai and B):
dXB / dt = ηB [cB(ωA1,B XA1, … , ωAk,B XAk) XB] (2)
XB(t+t) = XB(t) + ηB [cB(ωA1,B XA1(t), … , ωAk,B XAk(t)) XB(t)] t
concept
representation
explanation
State values
over time t
Y(t)
At each time point t each state Y in the
model has a real number value in [0, 1]
Single causal
impact
impactX,Y(t)
= X,Y X(t)
At t state X with connection to state Y has an
impact on Y, using connection weight X,Y
Aggregating
multiple
impacts
aggimpactY(t)
= cY(impactX1,Y(t),…, impactXk,Y(t))
= cY(X1,YX1(t), …, Xk,YXk(t))
The aggregated causal impact of multiple
states Xi on Y at t, is determind using
combination function cY(..)
Timing of the
causal effect
Y(t+t) = Y(t) + Y [aggimpactY(t) - Y(t)] t
= Y(t) + Y [cY(X1,YX1(t), …, Xk,YXk(t)) - Y(t)] t
The causal impact on Y is exerted over time
gradually, using speed factor Y; here the Xi
are all states with connections to state Y
One option for the combination functions for modelling the aggregated impact of
multiple states on another is the scaled sum function:
        (3)
Usually a normalised scaled sum is used: the value of is the sum of all incoming
weights ωXi,Y. In cases that these connection weights change over time an adaptive
version of the scaled sum can be used:
        (4)
where λ(t) is the sum of all incoming weights ωXi,Y(t) at t. This version was used to
model social contagion in the work reported here. Another function used for social
contagion in this research, is the advanced logistic sum function:
    
  
(1+e-στ) (5)
where is the steepness factor and log the logistic threshold.
Secondly, the principle of homophily describes the change of connection weights
between the states XA and XB of two persons A and B; e.g., [8]. According to this prin-
ciple, when the values of two nodes are similar, the connection between them be-
comes stronger (represented by a higher connection weight). Conversely, the lower
the similarity between the (values of) the two nodes, the smaller their connection
weight. In Fig. 1, the homophily principle is depicted by the striped arrows. Numeri-
cally, this principle can be represented as follows:
ωA,B(t+t) = ωA,B(t) + ηA,B [cA,B(XA(t), XB(t), ωA,B(t)) − A,B(t)]t (6)
A,B /dt = ηA,B [cA,B(XA, XB, ωA,B) − A,B ]
in which XA and XB represent the states of person A and person B.
In the current paper, the combination function cA,B(V1,V2,W) used for the homophi-
ly principle is the following:
slhomhom.(V1,V2,W) = W + W (1-W) (hom - |V1 V2|) (7)
Fig. 1: A conceptual representation of the homophily principle.
3 Simulations of example scenarios
In this section the adaptive network model is described for three scenarios. In the first
example Scenario 1 for this model, a social network of 10 nodes is used, consisting of
three communities of three or four nodes. Each community contains one node that is a
so called ‘bridge node’, which has connections to the two bridge nodes of the other
A,B
D,C
A,D
XA
XB
XC
XD
communities. Within the three communities, there is maximal connectedness: all
nodes are connected to the other nodes within the community. As the connections
represent social interactions, they are all assumed to be bidirectional. All connections
are presumed to be relatively strong, so all connection weights were set initially to
0.8. The state values represent the individual’s identification with the queer culture,
with 0 being minimal identification and 1 being maximal identification. The initial
values of identification with the queer culture are assumed to be spread evenly on the
spectrum of 0.1 to 1, as depicted in Fig. 2. For social contagion, in this scenario the
adaptive normalised scaled sum function was used, with a dynamic scaling factor of
the sum of all the connection weights per state at that time. Table 2 shows the values
used for the parameters. In this scenario, two simulations were carried out using two
different state speed factors. The simulation of this model shows a classic example of
the interplay of social contagion and homophily; sometimes also called co-evolution
[5, 18]. In Scenario 1.1, a speed factor of 0.2 was used for all states, and the three
communities all converged to their own equilibrium value. The connection weights of
the within-community connections converged to 1, while the weights of the bridge
connections converged to 0. This result is illustrated in Fig. 3.
Fig. 2: Left hand: conceptual representation of the network in Scenario 1. Middle: for Scenario
2. Right hand: for scenario 3. Each node represents an individual, with the initial state value
illustrated in the node. Each line represents a bidirectional connection. The three communities
are labelled with different colors.
Table 2: The parameters and their values used in the simulation of Scenarios 1 and 2. Except
for the initial connection weights mentioned above, all parameters are equal for all states and
connections. ( = connection weight, slhom = simple logistic homophily function, alogistic =
advanced logistic function as defined above)
Parameters Scenario 1
Values
Parameters Scenarios 2 and 3
Values
state speed factor Y
0.2/0.8
state speed factor Y
0.2
speed factor
0.5
speed factor
0.5
slhom threshold factor hom
0.1
slhom threshold factor hom
0.08
slhom amplification factor
8.0
slhom amplification factor
8.0
alogistic steepness factor
2.5
alogistic threshold factor log
0.18
The final example Scenario 3 concerns a fully connected network in which each node
is connected to every other node. Fig. 7 shows a conceptual representation of this
network. Again, the advanced logistic function is used to model social contagion be-
tween the nodes, and the simple homophily function alters the connection weights
over time. This scenario is somewhat more life-like than the previous examples, as it
is reasonable to assume that, in a group of 10 people, every person knows all others to
some degree. All initial connection weights are set to 0.8. The remaining parameter
values are shown in Table 4. In Scenario 1.2, when the state speed factor of 0.8 was
used, two of the communities first converge within themselves, and then converged to
a shared equilibrium state value. The third community converged to its own equilibri-
um value. The bridge connection between state 3 and 4 now converged to 1 instead of
0. This shows that the effect of the social contagion function is now quicker than in
simulation 1.1, and influences the homophily of the connection weights. This simula-
tion is depicted in Fig. 4.
Note that in case that no homophily principle is applied but only the social conta-
gion, according to Theorems 3 and 4 in [15] for this socalled strongly connected net-
work using normalised scaled sum combination functions (which are strictly mono-
tonically increasing and scalar-free [15]) all states will converge to the same value,
and this value lies between the minimal and maximal initial state value. The emer-
gence of communities is a result of the homophily, and the faster the contagion in
comparison to the homophily principle, the lower the number of communities that
emerge, as can be seen here in Figs. 3 and 4.
Fig. 3: A graphical representation of the state values in the simulation of Scenarios 1.1 resp. 1.2
for state speed factors 0.2 resp. 0.8. The y-axis represents the state values, while the x-axis
represents time. Every community converges to its own equilibrium value. Social contagion is
quicker in Scenario 1.2 than in Scenario 1.1. This leads to the fact that the first two communi-
ties converge towards each other. The third community still converges to its own equilibrium
In the example Scenario 2, the number of nodes and their connectedness is equal to
the first scenario. However, the initial values are range from 0.1 to 0.4 as depicted in
Fig. 5, and this time the advanced logistic function is used for social contagion. Table
3 shows the values of the parameters used in the simulation of this scenario. The re-
sults of this scenario shows that, when using the advanced logistic function to model
social contagion, the equilibrium values can end up higher or lower than any initial
values of the states, in contrast to what holds for normalised scaled sum functions; see
[15]. Most of the states converge to an equilibrium of 1 or 0.982, whereas all the ini-
tial values lay between 0.1 and 0.4. This may represent a real-life process in which a
sentiment is strengthened and amplified beyond its original level, because it is shared
with others.
Yet, the equilibrium values of state 5 and 6 (initial values 0.2 and 0.3) converge to
a different, much lower equilibrium of 0.415. The connection weights of all initial
connections converge to 1, except for those of state X4 with X5 and X6. This shows
that: 1) when using the advanced logistic function in combination with homophily, the
initial value of a state does not necessarily determine in which equilibrium it ends up;
and 2) that this combination function can strongly influence the connectivity in a net-
work, possibly leading to the change or dissolvement of communities over time. Thus,
this simulation demonstrates the phenomenon of segregation within a network (and in
a specific community within the network) regardless of a similarity in initial values of
the states. Fig. 6 shows these results in a graphical representation.
Fig. 6: A graphical representation of the state values in the simulation of Scenarios 2 and 3. The
y-axis represents the state values, while the x-axis represents time. The initial values of all
states range between 0.1 and 0.4, while most of the values converge to 1 or 0.982. However,
state 4 and 5 converge to 0.415.
Again, the simulation resulted in the separating behavior into two groups, as is
clearly visible in Fig. 8. The seven states with the highest initial values all converge to
a value of 0.8, spiraling past their original values. The remaining three states converge
to a value of 0.036, well below their initial values. The connection weights within the
groups all converge to 1, while the inter-group connections end up with a weight val-
ue of 0. This simulation not only shows separating behavior of one group into two
communities, as with the previous scenario. It also hints at a notion of extremism, in
which the two groups increasingly push each other off. This may be comparable to the
process of ‘othering’, in which the shaping of an identity depends on the supposition
with other people’s behavior or convictions. This combined with group behavior can
then lead to polarization, which can be observed in many social and political situa-
tions.
4 Model verification by Mathematical Analysis
In order to verify the model, first a mathematical analysis of stationary points was
performed, in particular for the third scenario. A stationary point of a state Y at time t
occurs when dY(t)/dt = 0. A stationary point of a connection weight at time t occurs
when d(t)/dt = 0. The network model is in an equilibrium at t when all states and all
connection weights have a stationary point at t. As described in Section 2, in a tem-
poral-causal network model the differential equation for all states is: dY(t)/dt = ηY
[aggimpactY(t) Y(t)]. As all speed factors in the model are nonzero, all stationary
points must follow the criterion: aggimpactY(t) = Y(t) also formulated as: in a tem-
poral-causal network model there is a stationary point for state Y at t if and only if ηY
= 0 or cYX1,YX1(t), …, ωXk,YXk(t)) = Y(t).
From the modelled data in Scenario 3, stationary points were gathered from several
states, and the aggregated impact at that time was calculated per state using the ad-
vanced logistic function described earlier. If the state values and the calculated aggre-
gated impact are equal, the stationary point equation above is fulfilled. This mathe-
matically verifies the model. The results are presented in Table 5. As appears in the
table, the deviations between the observed state values and the calculated aggregated
impact on that state at that time are very low. This indicates that the model does what
is expected; it calculates the expected state values with high precision.
Table 5: an overview of stationary point values and the aggregated impact values in the simula-
tion of Scenario 3. States 4 to 7 did not have a temporary stationary point at the considered time
interval. The bottom row shows the deviation between these values.
When the stationary point equation (9) mentioned above applies to all network
states and connection weights at a single time, the model is in equilibrium. In the third
scenario, the model appears to be in equilibrium at t = 300. The state values at this
time were read, and the aggregated impact at that moment was calculated per state.
The results are presented in Table 6. As is visible in the table, the deviation between
the state values and the aggregated impact of all states at t = 300 is very low, indicat-
ing again that the model calculates the state values in a proper way with high preci-
sion.
Additionally, the dynamic connection weights were analyzed. Recall the following
combination function for the homophily principle (7):
slhom.(V1,V2,W) = W + W (1-W) (hom - |V1 V2|) (10)
The stationary point criterion of slhomhom.(V1,V2,W) = 0 provides the equation:
W (1-W) (τhom - |V1 V2 |) = W (11)
meaning that
|V1 V2 | = τhom or A,B = 0 or A,B = 1 (12)
However, the solution |V1 V2 | = τhom turns out non-attracting, eliminating this
solution as a possible stationary point in the simulation. This corresponds to the con-
nection weight values at t = 300 in the simulation of Scenario 3: all connections even-
tually are either 1 or 0. These connections define the two clusters that appeared, with
full connectivity within the communities, and no connection to nodes outside the
community.
In a wider context such an analysis of limit behaviour for some classes of ho-
mophily combination functions has been presented in [16]. The above analysis fits in
that more general approach. In addition to the 0 or 1 values as limit for connection
weights, one of the results is that independent of the size of the network there can be
at most 1+1/hom groups; see [16], Theorem 1a). Indeed, the actual number of the
groups in the simulations is less than that predicted maximal number.
State
X1
X2
X3
X8
X9
X10
Time point
3.95
3.55
2.10
4.65
3.10
2.55
State value
0.387
0.397
0.397
0.261
0.197
0.145
Aggregated impact
0.388
0.396
0.397
0.262
0.197
0.145
Deviation
0.001
0.001
0
0.001
0
0
Table 6 The state values in the simulation of Scenario 3 at t = 300.
5 Validation using empirical data
1
To validate the proposed model, we acquired a data set for which we set up a ques-
tionnaire that participants had to fill out online. The questionnaire consisted of some
general introductory questions like age, gender, education and most importantly sexu-
al orientation. Secondly, the participant had to indicate to what extent they agreed to
statements (using a Likert scale, from 1 till 5, 1 indicating “strongly disagree” and 5
indicating “strongly agree”). The first 10 questions related to how involved they are in
the queer community now. The second 10 questions related to how involved they
were in the queer community 5 years ago. A final 10 questions related to how much
social hardship the participant experienced in their youth, based both on a general
level and in relation to their sexuality.
To explore the suitability of our data, we used SPSS to perform a statistical analy-
sis regarding the following hypotheses: 1) that queer people would generally score
higher on involvement in the queer scene than straight people, 2) that queers would be
involved more in the community now than 5 years ago, and finally 3) that hardship
would make up for explaining this difference between involvement in the queer
community now versus 5 years ago.
By exploring the differences in scores on involvement in the queer scene between
queer people and straight people, two of our hypotheses were confirmed: queer people
generally score higher on involvement in the queer scene than straight people, and
queers are involved more in the community now than 5 years ago. Another analysis
was needed to check the assumption that differences in scores were dependent on the
hardship that people experienced by verifying a (possible) a correlation of the scores
with the scores on the hardship questions. However, a Pearsons correlation r of 0,056
indicated no correlation for the variables. This refutes the hypothesis that the increase
in identification with the queer community is mediated by the degree of social hard-
ship experienced when young.
Next the empirical data were reshaped in a format compatible with the format of
the simulation outcomes, so that parameter tuning would provide the best possible
solution for the model and the empirical data to fit together. Scenario 3 was chosen
for the parameter tuning as it would be the best possible fit for real-life scenarios: as a
case where each individual knows each other individual resembles a real-world sce-
nario most. The connection weights for the model were set under the assumption that
some segregation was already in place: straight people were set to a connection
1
A more detailed account of this section can be found at URL
State
X1
X2
X3
X4
X5
X6
X7
X8
X9
X10
State value
0.802
0.802
0.802
0.802
0.802
0.802
0.802
0.040
0.036
0.038
Agg. impact
0.803
0.802
0.803
0.803
0.802
0.801
0.802
0.038
0.037
0.037
Deviation
0.001
0
0.001
0.001
0
0.001
0
0.002
0.001
0.001
weight of 0.8 to other straight people, knowing mostly other straight people and a
connection weight of 0.3 to queers. For queers it was the other way around, setting
connection weights to other queers at 0.8 and to straights at 0.3. This leaves some
parameters of the model to still be tuned: the speed factor ηX for each state X, the
steepness σ of the alogistic combination function, the threshold τlog of the alogistic
combination function, the threshold τhom,X,Y of the simple linear homophily, and the
amplification factor αX,Y of the simple linear homophily.
The way we estimated these parameters was through exhaustive search. The ex-
haustive search method is a problem-solving technique in which all possible candidate
solutions for parameter values are investigated on how well they make the model fit
to the data. Because testing each parameter with a grain size of 0.05 would lead to
combinatorial explosion, the way to execute the exhaustive search was by iterative
refinement, starting with larger grain sizes, for example 0.5 or 0.1 (depending on the
parameter) and narrowing down the grain sizes until the model fits the empirical data
best. Speed factors, steepness and thresholds were investigated with grain size 0.1,
and then in a second phase tuned with a grain size of 0.05.; however amplification
was investigated with a grain size of 1, and then further tuned with a grain size of 0.5.
The values found are = 0.53, τlog = 0.2, τhom = 0.1, = 3.5, and ηX1 = 0.1, ηX2 = 0.1,
ηX3 = 0.1, ηX4 = 0.35, ηX5 = 0.3, ηX6 = 0.3, ηX7 = 0.3, ηX8 = 0.3, ηX9 = 0.3, ηX10 = 0.3.
Simulation outcomes for the tuned parameters are depicted in Fig. 7. An overview of
the remaining errors is shown in Table 7. Naturally, some variances exist between the
proposed model and the empirical data. Using the Root Mean Square to calculate the
differences between the values of the model equilibria and equilibria of the empirical
data (the final state of involvement in the queer community), this difference should be
kept to a minimum in tuning the parameters. The achieved results of the calculated
average (absolute) deviation (0.2495) and Root Mean Square (0.3492) are depicted in
Table 7, last column.
Fig. 7 Simulation modeling the empirical data. The X-axis represents time and the Y-axis
represents the identification with the community.
Table 7 Calculated Error: Average absolute deviation and Root Mean Square
6 Conclusion
This paper investigated the interplay or co-evolution of the social contagion principle
and the homophily principle in their application in an adaptive temporal-causal net-
work model. Both principles were modeled, and applied in three model scenarios.
State values represented personal convictions, while the connections between states
represented real-life social interaction, leading to influencing behavior.
The first scenario used an adaptive normalised scaled sum function to model so-
cial contagion. It showed the segregation behavior of a group of ten people into either
two or three communities, depending on parameters such as speed factors, threshold
factors and steepness factors. The second scenario used the advanced logistic sum
function, and showed a separation into two communities, that were not entirely de-
fined by the initial grouping of connections. Furthermore, it demonstrated the spiral-
ling of values beyond the range of initial values: a pattern that is not achievable with
linear functions such as scaled sum functions. This pattern is sometimes called ‘emo-
tion amplification[3], where emotions or opinions are amplified through sharing
them.
The simulation of the third scenario also used the advanced logistic sum function,
with the addition of connections to all other nodes in the network. This scenario is
slightly more life-like. This scenario showed the segregation into two groups, and the
amplifying behavior as discussed in Scenario 2. Furthermore, the third scenario
showed a pattern of polarization, in which the values of the two groups increasingly
move apart. In social terms, this can be compared to group identification, in which the
more the other team disagrees with you, the stronger your opinion gets. It also shows
signs of a process called ‘othering’, in which an individual’s or group’s identity
strongly depends on what they are not. These phenomena can be observed in many
social situations, with the political system of the United States being a classic exam-
ple. This process may well contribute to segregating behavior, including political
‘echo chamber’ that is online social media [12], or the taking shape of subgroups or
communities.
The final section of this paper described our attempt to gather empirical data re-
garding the shaping of the queer community, and the segregating behavior that it re-
lates to. Before using the data in the network model, a statistical analysis showed that
State
X1
X2
X3
X4
X5
X6
X7
X8
X9
X10
Total
Empirical Equilibrium
0.357
0.597
0.583
0.601
0.694
0.718
0.778
0.856
0.745
0.88
Model Equilibrium
0
0
0
0
0.747
0.747
0.747
0.747
0.747
0.747
Absolute Deviation
0.357
0.597
0.583
0.601
0.053
0.029
0.031
0.109
0.002
0.133
0.2495
Average deviation
Square of Deviation
0.127
0.356
0.340
0.361
0.003
0.001
0.001
0.012
0.000
0.018
1.219
SSR
Root Mean
Square
0.3
492
12
there was a significantly higher identification with the queer community of sexual
queers than heterosexual participants. It also revealed that queer people had a higher
number of queer friends, and showed a higher increase in identification with the queer
community than heterosexuals. The hypothesis stating that the level of identification
increase is mediated by social hardship experienced in youth, was rejected. The em-
pirical data was transformed to fit the model (based on Scenario 3), and an exhaustive
search tuning method was performed in order to tune the parameters to best fit the
empirical data. The lowest average linear deviation was found to be 0.386. This is
relatively high, which can partly be attributed to the fact that the empirical data used
an average of scores, leading to values that lie relatively close to one another but are
still significantly different. In the model, however, every state represented the average
of a group of 5 or 6 people. Moreover, the combination of the advanced logistic sum
function and the simple homophily function has a polarizing tendency (as described in
scenario 2), which was not clearly visible in the empirical data.
The current research may be improved by using more than 10 nodes, preferably 50
or more. Not only would this overcome the limitation of having to group participants
together and losing their unique trends, but it also might be expected that the interplay
between 50 nodes is vastly different from that of 10 nodes, in the way that a class-
room with fifty children acts different than one with ten.
Additionally, the empirical data regarding the correlation between youth social
hardship and connectedness to the queer community did not show a correlation (or
even a trend). This may be due to the survey used, in which only 10 questions were
focused on the general social hardship, while a focus on sexuality-related hardship
would have been more useful. Secondly, the survey answers are strongly constrained
by the snowball effect that was used for gathering the data: many people who filled in
the questionnaire knew others, which results in bias when attempting to study the
formation of communities.
Finally, the study would be highly improved if the individual connections between
people would be investigated over time. Now, questions about the number of queer
people in the individual’s network were used to approximate the average effect, while
the combination of using the real individual interactions in a network of 50 nodes
would give deeper insight in the workings of human group behavior.
Overall this research might be considered as a step in the direction of understand-
ing more about segregation and polarizing behavior. Especially the adaptiveness,
using homophily, is indispensable for the future of creating temporal-causal models of
human interactions and their consequences.
References
1. Batson, C.D., Ahmad, N., and Tsang, J.-A.: Four motives for community involvement.
Journal of Social Issues 5(8.3): 429-445 (2002)
2. Beemyn, B.: Creating a place for ourselves: Lesbian, gay, and bisexual community histo-
ries. Routledge (2013)
3. Bosse, T., Duell, R., Memon, Z.A., Treur, J., and Wal, C.N. van der: Agent-Based Model-
ing of Emotion Contagion in Groups. Cogn Comput 7, 111136 (2015)
13
4. Harcourt, J.: Sexual identity distress, social support, and the health of gay, lesbian, and bi-
sexual youth. In Current Issues in Lesbian, Gay, Bisexual, and Transgender Health (pp.
97-126). Routledge. (2013)
5. Holme, P., and Newman, M.E.J.: Nonequilibrium phase transition in the coevolution of
networks and opinions Phys. Rev. E 74(5), 056108 (2006)
6. Kuipers, B.J.: Commonsense reasoning about causality: Deriving behavior from structure,
Artificial Intelligence, 24, 169-203, (1984).
7. Kuipers, B.J., and J.P. Kassirer: How to discover a knowledge representation for causal
reasoning by studying an expert physician. In: Proceedings Eighth International Joint
Conference on Artificial Intelligence, IJCAI’83, Karlsruhe. William Kaufman, Los Altos,
CA, (1983).
8. McPherson, M., Smith-Lovin, L., & Cook, J.M.: Birds of a feather: Homophily in social
networks. Annual review of sociology 27(1), 415444 (2001)
9. Meyer, I.H.: Minority stress and mental health in gay men. Journal of health and social be-
havior 36, 38-56 (1995)
10. Meyer, I.H.: Prejudice, social stress, and mental health in lesbian, gay, and bisexual popu-
lations: conceptual issues and research evidence. Psychology of Sexual Orientation and
Gender Diversity 1(S), 3-26 (2013)
11. Pearl, J.: Causality. Cambridge University Press (2000).
12. Sunstein, C.R.: # Republic: Divided democracy in the age of social media. Princeton Uni-
versity Press (2018)
13. Treur, J.: Dynamic modeling based on a temporalcausal network modeling approach. Bi-
ologically Inspired Cognitive Architectures 16, 131-168 (2016a)
14. Treur, J.: Network-Oriented Modeling: Addressing Complexity of Cognitive, Affective
and Social Interactions. Springer Publishers (2016b)
15. Treur, J.: Relating Emerging Network Behaviour to Network Structure. In: Proc. of the 7th
International Conference on Complex Networks and their Applications, ComplexNet-
works'18, vol. 1. Studies in Computational Intelligence, vol. 812, pp. 619-634. Springer
Publishers (2018a)
16. Treur, J.: Relating an Adaptive Social Network’s Structure to its Emerging Behaviour
Based on Homophily. In: Proc. of the 7th International Conference on Complex Networks
and their Applications, ComplexNetworks'18, vol. 2. Studies in Computational Intelli-
gence, vol. 813, pp. 341-356. Springer Publishers ( 2018b)
17. Treur, J.: The Ins and Outs of Network-Oriented Modeling: from Biological Networks and
Mental Networks to Social Networks and Beyond. Transactions on Computational Collec-
tive Intelligence 32, 120-139. Based on Keynote Lecture at the 10th International Confer-
ence on Computational Collective Intelligence, ICCCI’18 (2019)
18. Vazquez, F.: Opinion dynamics on coevolving networks. In: A. Mukherjee, M.
Choudhury, F. Peruani, N. Ganguly, and B. Mitra (eds): Dynamics On and Of Complex
Networks, Volume 2, Modeling and Simulation in Science, Engineering and Technology,
pages 89107. Springer New York (2013)
... Several examples of combination functions in adaptive temporal-causal networks are used in literature such as [3,7,8,11]; see Table 3 for some of them. The first is the identity id(.) for a single state impacting another state. ...
Conference Paper
Full-text available
Alcohol consumption among young adolescents is problematic as health implications and behavioural changes are common consequences. Another problematic factor among young adolescents is the amount of delinquencies committed. In this paper an adaptive social agent network model using friendship relationships as predictor for alcohol consumption and amount of delinquencies committed is explored. The proposed agent network model was empirically validated using classroom data on young adolescents gathered by Knecht in Dutch schools. The agent network model is second-order adaptive and applies a bonding by homophily adaptation principle with adaptive adaptation speed describing clustering in friends networks based on the two aforementioned factors respectively .
Chapter
Full-text available
In this chapter, it is analysed how emerging behaviour in an adaptive network for bonding based on similarity (homophily) can be related to characteristics of the adaptive network’s structure, which includes the structure of the adaptation principles incorporated. In particular, this is addressed for adaptive social networks for bonding based on homophily and community formation. To this end, relevant characteristics of the reified network structure (including the adaptation principle) have been identified, such as a tipping point for similarity as used in the aggregation for homophily. Applying network reification, the adaptive network characteristics are represented by reification states in the extended network, and adaptation principles are described by characteristics of these reification states, in particular, their connectivity characteristics (their connections) and their aggregation characteristics (in terms of their combination functions). According to this network reification approach, as one of the results, it has been found how the emergence of communities strongly depends on the value of this similarity tipping point. Moreover, it is shown that some characteristics entail that the connection weights all converge to 0 (for persons in different communities) or 1 (for persons within one community). This is Chapter 13 of the book https://www.researchgate.net/publication/334576216.
Article
Full-text available
In this paper it is analysed how community formation in an adaptive network for bonding based on similarity (homophily) can be related to characteristics of the adaptive network's structure, which includes the structure of the adaptation principles incorporated. In particular, this is addressed for adaptive social networks for bonding based on homophily. To this end, relevant properties of the network and the adaptation principle have been identified, such as a tipping point for similarity. As one of the results, it has been found how the emergence of communities strongly depends on the value of this tipping point. Moreover, it is shown that some properties of the structure of the network and the adaptation principle entail that the connection weights all converge to 0 (for persons in different communities) or 1 (for persons within one community). Published in Applied Network Science Journal, 2019, Special Issue on Community Structure in Networks, Guest editors Gergely Palla, Hocine Cherifi, Boleslaw K. Szymanski.
Conference Paper
Full-text available
In this paper it is analysed how emerging behaviour of an adaptive network for bonding based on homophily can be related to characteristics of the adaptive network's structure , which includes the structure of the adaptation principles incorporated. Relevant properties of the network and the adaptation principle have been identified, such as a tipping point for homophily. As one of the results it has been found how the emergence of clusters strongly depends on the value of this tipping point. Moreover, it is shown that some properties of the structure of the network and the adaptation principle entail that the connection weights all converge to 0 (for states in different clusters) or 1 (for states within one cluster).
Conference Paper
Full-text available
Emerging behaviour of a network is usually considered to be a consequence of the network’s structure, although it may be challenging to find out how this relation between structure and behaviour exactly is. In this paper a number of results are presented on how certain identified properties of network structure determine network behaviour, in particular what limit behaviour emerges. These network structure properties include in how far other states of the network are reachable from a given state, and properties of combination functions used to aggregate multiple impacts on a single state.
Article
Full-text available
Network-Oriented Modeling has successfully been applied to obtain network models for a wide range of phenomena, including Biological Networks, Mental Networks, and Social Networks. In this paper it is discussed how the interpretation of a network as a causal network and taking into account dynamics in the form of temporal-causal networks, brings more depth. The basics and the scope of applicability of such a Network-Oriented Modelling approach are discussed and illustrated. This covers, for example, Social Network models for social contagion or information diffusion, adaptive Mental Network models for Hebbian learning and adaptive Social Network models for evolving relationships. From the more fundamental side, it will be discussed how emerging network behavior can be related to network structure. This paper describes the content of my Keynote lecture at the 10th International Conference on Computational Collective Intelligence, ICCCI'18.
Book
Full-text available
This book has been written with a multidisciplinary audience in mind without assuming much prior knowledge. In principle, the detailed presentation in the book makes that it can be used as an introduction in Network-Oriented Modelling for multidisciplinary Master and Ph.D. students. In particular, this implies that, although also some more technical mathematical and formal logical aspects have been addressed within the book, they have been kept minimal, and are presented in a concentrated and easily avoidable manner in Part IV. Much of the material in this book has been and is being used in teaching multidisciplinary undergraduate and graduate students, and based on these experiences the presentation has been improved much. Sometimes some overlap between chapters can be found in order to make it easier to read each chapter separately. Lecturers can contact me for additional material such as slides, assignments, and software Springer full-text download: http://link.springer.com/book/10.1007/978-3-319-45213-5
Article
Full-text available
This paper presents a dynamic modelling approach that enables to design complex high level conceptual representations of models in the form of causal-temporal networks, which can be automatically transformed into executable numerical model representations. Dedicated software is available to support designing models in a graphical manner, and automatically transforming them into an executable format and performing simulation experiments. The temporal-causal network modelling format used makes it easy to take into account theories and findings about complex brain processes known from Cognitive, Affective and Social Neuroscience, which, for example, often involve dynamics based on interrelating cycles. This enables to address complex phenomena such as the integration of emotions within all kinds of cognitive processes, and of internal simulation and mirroring of mental processes of others. In this paper also the applicability has been discussed in general terms, showing for example that every process that can be modelled by first-order differential equations, also can be modeled by the presented temporal-causal network modeling apporoach. A variety of example models that can be found in other papers illustrate the applicability of the approach in more detail.
Article
Full-text available
In this article the author reviews research evidence on the prevalence of mental disorders in lesbians, gay men, and bisexuals (LGBs) and shows, using meta-analyses, that LGBs have a higher prevalence of mental disorders than heterosexuals. The author offers a conceptual framework for understanding this excess in prevalence of disorder in terms of minority stress— explaining that stigma, prejudice, and discrimination create a hostile and stressful social environment that causes mental health problems. The model describes stress processes, including the experience of prejudice events, expectations of rejection, hiding and concealing, internalized homophobia, and ameliorative coping processes. This conceptual framework is the basis for the review of research evidence, suggestions for future research directions, and exploration of public policy implications.
Article
Full-text available
To avoid the development of negative emotion in their teams, team leaders may benefit from being aware of the emotional dynamics of the team members. To this end, the use of intelligent computer systems that analyze emotional processes within teams is a promising direction. As a first step towards the development of such systems, this paper uses an agent-based approach to formalize and simulate emotion contagion processes within groups, which may involve absorption or amplification of emotions of others. The obtained computational model is analysed both by explorative simulation and by mathematical analysis. In addition, to illustrate the applicability of the model, it is shown how the model can be integrated within a computational 'ambient agent model' that monitors and predicts group emotion levels over time and proposes group support actions based on that. Based on this description, a discussion is provided of the main contribution of the model, as well as the next steps needed to incorporate it into real world applications.
Article
Creating a Place For Ourselves is a groundbreaking collection of essays that examines gay life in the United States before Stonewall and the gay liberation movement. Along with examining areas with large gay communities such as New York, San Francisco and Fire Island, the contributors also consider the thriving gay populations in cities like Detroit, Buffalo, Washington, D.C., Birmingham and Flint, demonstrating that gay communities are truly everywhere. Contributors: Brett Beemyn, Nan Alamilla Boyd, George Chauncey, Madeline Davis, Allen Drexel, John Howard, David Johnson, Liz Kennedy, Joan Nestle, Esther Newton, Tim Retzloff, Marc Stein, Roey Thorpe.