You are on page 1of 9

Application of IFS for Conflict Resolution

Modeling and Agent Based Simulation

Shpend Ismaili1 and Stefka Fidanova2


1
University of Tetovo, Macedonia
shpend.ismaili@unite.edu.mk
2
Institute of Information and Communication Technology
Bulgarian Academy of Science, Bulgaria
stefka@parallel.bas.bg

Abstract. The complexity of social systems and interactions between


its members is difficult to be represented. In tis paper we focus on conflict
situations that can happen during a protest. Our model is based on soft-
ware agents. In our case there are three types of agents: active agents;
peaceful agents and police officers. We will model the interactions be-
tween agents and their consequences. Different events will be simulated
with fuzzy methods and intutionistic relations. Thus will be determine
the level of similarity between the agents.

1 Introduction

Social systems are very complex and difficult to be predicted due to the complex
interactions inside the system and with the environment outside the system. A
social system consist of individuals who interact with each other, evolve in an
autonomous manner motivated by their beliefs, personal goals and the circum-
stances of the social environment. We will model these complex systems with
autonomous software agents that interact with each other and their environ-
ment. The agent simulates the individual in a social system [10]. Studies on real
estate agents can help in the analysis of behavior and the evolution of the social
system. Modeling and simulation is done in controlled environments and provide
a platform for empirical study of social systems. However, there are some limita-
tions in the procedure for modeling such systems. We need to focus on modeling
specific social processes.
This paper presents a conflict situation that occurs due to dissatisfaction with
the central government which may be due to the illegitimacy of the government,
complaints, greed or other reasons. There are models presented with Agents
Based Modeling (ABM) that are proposed: for modeling civilian violence [8, 9],
confrontations between two opposing groups [12], rebellion [6] as well as wars
[11]. In our model three types of agents are specified: active civilians who are part
of the protesters, peaceful civilians who do not have their status defined, and
police officers guarding government institutions and trying to calm the angry
crowd. We will focus on the interactions of active agents with peaceful agents
2 Sh. Ismaili, S. Fidanova

and police agents with active and peaceful agents. Active agents communicate
with peaceful civilians and want to persuade them to join the crowd, while
police officers communicate with peaceful agents to stay calm and if they do not
join the crowd, they protect them from possible violent activists if the situation
escalates.
The core of the model that we present are agents. To reach a model that will
model the event more crucially is to describe what is closer to human life and its
behavior in situations when they are in the crowd. It is therefore very important
that it is as precisely as possible to model their attributes, which is decisive
for describing as closely as possible the real life of man and behavior in the
social turbulence, protests in the concrete case. In situations of protest resulting
from dissatisfaction with the central government, peaceful civilian citizens are
neutral participants in the society that is settled in the midst of troubles and
difficulties. They do not pose a threat to the central government, but from time
to time they react to external or internal stimuli. They are silent and quiet
laws, but they become active if the conditions are favorable, in order to express
revolt and irritation in public. Police officers are keeping their routine through
the involvement of activists in the jail and through strategies that determine
the success of management and control of violence. The main attributes that
contribute to the protest are: insufficient, mournful and greed, the legitimacy of
power, and the danger of arrest by the police.
The rest of the paper is organized as follows. In Section 2 is introduced the
fuzzy logic and the agents based modeling. In Section 3 are described the intu-
itionistic relations in agent based modeling context. In section 4 are simulated
different events with fuzzy methods. In section 5 is determine the similarity be-
tween the agents. In section 6 some concluding remarks and future works are
written.

2 Intuitionistic Fuzzy Logic

Over the last years, there has been an increase in interest in social sciences
to apply fuzzy logic. For example, in some ABMs, agents make decisions by
fuzzy rules; ”Fuzzy controls” or ”fuzzy agents”, which are expert systems that
are based in the rules ”if → then” where the premises and the conclusions are
unclear. In contrast to traditional agent-based models, where these are in a
complete way certain which is the forgiveness of the real individuals, represented
by agents, the fuzzy agents also take into account the stochastic complement of
human behavior. In this paper we try to apply strategies, selected from the
iterative dilemma of the prisoners [4], using IF logic rules [1–3] for making the
decision [13]. In this way in simulation based on a two-player game, we can use
fuzzy strategies when analytic solutions do not exist or are very difficult to get
(because agents use fuzzy strategies, for example, ”If I think my opponent will
choose action x, I will select the action y) [17].
Let U is not empty set. An intuitionistic fuzzy set A from U has the following
form:
Title Suppressed Due to Excessive Length 3

A = {hx, µA (x), νA (x)i|x ∈ E}, (1)


where the functions µA (x), νA (x) : U → [0, 1] defines in a corresponding
manner the degree of membership or non-membership of an element [3]. We
have that πA (x) = 1 − µA (x) − νA (x) and is called the index of uncertainty in
Intuitionistic Fuzzy Set (IFS) or hesitation of x in A. πA (x) is the degree of
uncertainty of x ∈ U of IFS A and πA (x) ∈ [0, 1] and 0 ≤ πA (x) ≤ 1 for every
x ∈ U . πA (x) is a lack of knowledge of whether it belongs to IFS A or not.
Intuitionistic Fuzzy (IF) logic is useful in environments with obscure and
unreliable (unsafe) information about some aspects of studying human society.
Many features of a person do not have a clear boundary or this boundary depends
on the interpretation or context, for example how long or how nice a person is,
but this can be represented by an IFS that describes the person himself and
gives them a degree of membership of certain characteristic. For an IFS that
is composed of several characteristics, it is sometimes necessary to model and
act with algebraic operations over them. So it is necessary to be modified, to
perform fuzification step by step. By using the IF logic, we can visibly increase
the accuracy of the model and their evolution. Before we use the IF operators, we
first need to fuzzy variables, which represent the characteristics (attributes) of
the individuals who participate in the crowd, ”active”, ”policemen”, ”peaceful”
agents who are used, or in a formal way we define IFS for these variables. We need
to use different operators to define them algebraically or logically, generalized
in classical operations: union, intersection, negation, or complement that are
defined in certain sets. First of all, we need to define the operators intuitionistic
t-norm and t-conorm, which we will use later. To define t-norm or t-conorm
(s-norm) we define L∗ [7].
Let:
L∗ = {(x1 , x2 ) ∈ [0, 1] × [0, 1]|x1 + x2 ≤ 1}.
Relation ≤L∗ in L∗ is defined as:

∀(x1 , x2 ), (y1 , y2 ) ∈ L∗ ,
(x1 , x2 ) ≤L∗ (y1 , y2 ) <=> x1 ≤ y1 and x2 ≥ y2
The relation ≤L∗ is partially ordered in L∗ and the ordered couple (L∗ , ≤L∗ )
is a complete lattice with the smallest element 0L∗ = (0, 1) and with a the largest
element 1L∗ = (0, 1)
Thus the descriptive definitions according to [7] will be:
One t-norm to one (L∗ )2 - T is a relation that satisfies the following condi-
tions:

– (∀x ∈ L∗ )(T (x, 1L∗ = x)) border conditions;


– (∀(x, y) ∈ (L∗ )2 )(T (x, y) = T (y, x)) commutativity;
– (∀(x, y, z) ∈ (L∗ )3 )(T (x, T (y, z)) = T (y, T (x, z))) associativity;
– (∀(x, x0 , y, y 0 ) ∈ (L∗ )4 )(x ≤L∗ x0 and y ≤L∗ y 0 => (T (x, y) ≤L∗ T (x0 , y 0 )))
monotony
4 Sh. Ismaili, S. Fidanova

One S-norm to one (L∗ )2 - S is a relation that satisfies the following condi-
tions:

– (∀x ∈ L∗ )(T (x, 0L∗ = x)) border conditions;


– (∀(x, y) ∈ (L∗ )2 )(S(x, y) = S(y, x)) commutativity;
– (∀(x, y, z) ∈ (L∗ )3 )(S(x, S(y, z)) = S(y, S(x, z))) associativity;
– (∀(x, x0 , y, y 0 ) ∈ (L∗ )4 )(x ≤L∗ x0 and y ≤L∗ y 0 => (S(x, y) ≤L∗ S(x0 , y 0 )))
monotony

In the Intuitionistic theory, t-norm is used to model intersection between two


fuzzy sets and in equivalent way is modeled the union between two fuzzy sets.

A ∩T B(u) = T (A(u), B(u))


A ∪S B(u) = S(A(u), B(u))
The other operator we need is also negation, which is defined as:
A IF negation in L∗ is a relation N : L∗ − > L∗ , which satisfy: N (0L∗ ) = 1L∗
and N (1L∗ ) = 0L∗ . If N (N (x)) = x for all x ∈ L∗ , than N is called involu-
tion IF negation, similar the relation NS , which is defined as N (N (x1 , x2 )) =
(x1 , x2 ) for every (x1 , x2 ) ∈ L∗ is called standard negation. The operators ∧
and ∨ in (L∗ , ≤L∗ ) are related with ordering ≤L∗ and are defined respectively,
∀(x1 , x2 ), (y1 , y2 ) ∈ L∗ :

(x1 , x2 ) ∧ (y1 , y2 ) = (min(x1 , y1 ), max(x2 , y2 )),


(x1 , x2 ) ∨ (y1 , y2 ) = (max(x1 , y1 ), min(x2 , y2 )),
and x = y <=> x ≤L∗ yand y ≤L∗ x
In many cases when using IF logic, IF t-norms and IF t-conorms are used.

3 Intuitionistic Fuzzy relations in ABM context

We will define the IFS and IF relationships in the set U with agents, U =
{Individi }i=1,N to present different attributes, linguistic variables, and rela-
tionships between agents (their interactions). An analysis of the evolution of at-
tributes in human society is a case that is important for the agent-based model.
However, the abstraction of modeling with agents in some aspects is very sim-
plified and does not give social science scientists the interpretation of results in
the same conditions that the survey data is expressed in real life. Thus, some
simplifications also produce several divergences regarding the data we are study-
ing. These facts motivate the use of the IF logic to get models with agents closer
to the real system. In this way, ABM uses the IF logic in some aspects, such
as: relationships between agents, some different attributes that determine the
properties of agents, the function of similarity, the evolution of real conditions
of agents, etc.
Title Suppressed Due to Excessive Length 5

In the context of the theme in multi-agent systems (MAS), the interactions


between agents ”active” and ”peaceful” are determined, as well as ”policemen”
and ”peaceful”, in which in this interaction the active agent can ”persuade”
or ”not persuade” the peaceful agent to join the crowd (mass) as well as the
interactions between police and peaceful civilians, who try not to allow them to
join the crowd and keep them from violent agents. These ratios naturally have
the predisposition to become fuzzy, because if these relations are defined like
a Boolean where two agents can be convinced or not, it is unrealistic in real
life. We will try to use the model that can present these relations more closely
using one continuous process of ”conviction”. Because of this, this interaction is
determined as an IF relation, with a real degree of interactions between 0 and 1
for each pair of agents located in the neighborhood.
Let Ragent : U × U → [0, 1] be IF relation in a group of agents that gives
a degree of ”interaction (to convince)” between agents of different groups. This
IF ”persuasion” gives one degree of interaction in the interval [0, 1] for each pair
of agents in the neighborhood (who are able to communicate). Let Ind be an
agent in U . The set Agent(Ind) is defined as the set of all agents x ∈ U where
Ragent (Ind, x) is greater than 0. Every agent from a situation when ”encountered
by accident” another agent from other groups (another type of agents) who are
inside their vision is interacting. Some limitations can be introduced in this way
of definition in order to adapt to the needs of the context. The relation ”same
goals (disregard for the authorities, complaints ...)” is also fuzzy with the fact
that this relationship will be ”1” if there are much greater dissatisfaction and
”0” with agents close to the government and a level between 0 and 1 for others.
Thus, features that will be ”peaceful”, or remain ”peaceful” is modeled with IF
relationship Rkooperon : U × U → [0, 1].

3.1 Simulating events with Fuzzy Methods

Social systems are complex adaptive systems, then we will try to simulate an
event that is under development, at the moment we look at it. Once a decision
has been taken (for example, becoming active or remaining peaceful), then such
a decision can be defined as a classic relation: after a peaceful agent’s decision
will go into ”active” or remain ”peaceful”. We suggest to apply this classic
relationship Rf qinje : U × U → {0, 1} to use approximate reasoning and IF
relations. Surely, if we know the relation Rf qinje , and we also know a classical
set U, which is defined as:

1, ∃ ind2|Rf qinje (ind, ind2) = 1
Active(ind) = , (2)
0, otherwise

peacef ul(ind) = N OT 0 active0 (ind)AN D N OT 0 police0 (ind)


The attributes that will be presented here as variables that decide whether an
agent will make a decision about his status in the crowd (if there is displeasure
to the central government, whether the central government has no legitimacy,
6 Sh. Ismaili, S. Fidanova

the duration of the protest, and if the chances are to be arrested then, the risk
is low), all expressed by the parameters G-Appeals, L-legitimacy, GR-greed, Tf-
factor time, RA-risk, as defined above in the model [8]. One has been chosen
to be ”more compact”, since such compactness is defined as the combination of
what is the result of the harmfulness of the above mentioned attributes and the
likelihood of being misleading (according to the deficiencies of the variables it
is established whether these are peaceful , active or policeman). Even this is a
simplification of actual events and later it will use some ”coincidence” in this
process. This important information on how the interaction between the two
agents can be done can be taken as a conclusion from the IF relation that will
be defined as Rcompatible : U × U → [0, 1], which can be obtained using the
IFOWA aggregation operator , [18], and operations in the classical sets:

”active” : U → [0, 1]; ”peacef ull” : U → [0, 1]; ”police” : U → [0, 1]

and an IF rule for the conclusion, where the premise is conjuction from the
classical sets. Concerning the conclusion, we will use the IF Ordered Weighted
Averaging (IFOWA) [18], which represents the aggregation of multi-criteria pro-
cedures. By specifying the ordered weights w (whose sum will always result in
1) it is possible to change the shape of the aggregation.
We can formally determine the IF relation Rcompatible : U × U → [0, 1], using
the IF relation Rsimilarity : U × U → [0, 1] and Ragent : U × U → [0, 1] as in the
following relations:

Rcomptabile (Ind, Ind2) = IF OW A(Ragent (Ind, Ind2), Rsimilarity (Ind, Ind2)) =

= w1 ∗ Ragent (Ind, Ind2) ⊕ w2 ∗ Rsimilarity (Ind, Ind2)

for all Ind, Ind2 ∈ U where w1 + w2 = 1.


After calculating the IF relation Rcompatible , IF relation Rinteraction : U ×
U → [0, 1] can be calculated using the set of ”peaceful” U , as below:

Rinteraction (Ind, Ind2) = aktive(Ind) AN D peacef ul(Ind2)AN DRcomptable (Ind, Ind2)

For police agents we will have:

Rinteraction (Ind, Ind2) = police(Ind) AN D peacef ul(Ind2)AN DRcomptable (Ind, Ind2)

We emphasize that the AND conjunction is implemented by an IF t-norm.


This can be done many times due to the associative properties of IF t-norms.
So T (x, y, z) = T (T (x, y), z)
Title Suppressed Due to Excessive Length 7

3.2 Determine the Similarity Between Agents


To calculate fuzzy relations Rcomptable used to see the status of agents, we have
used the IF relationship Rsimilarity : U × U → [0, 1]. In MAS, the similarity is
modeled and implemented through an algorithm that collects more data that
compares the clearly defined attributes of agents. While defining IFS over these
variables (attributes) and fuzzing the operator of similarity in their base, we will
be able to have much greater accuracy in determining the similarity between
individuals. In addition, with these fuzzy sets we will be able to draw a conclusion
based on them. First IFS is determined based on the variables defined by each
agent. Although, if there are some variables that can not be fuzzed, more of them
will be defined in the IFS. Now, with the formation of an IFS with attributes that
characterize the particular agent, with them we can determine the IF similarity
that generalizes the classical relation of equivalence. Thus, using it, we define
the similarity as T-indistinguishability (as in [16]) which generalizes the classical
relation of equivalence. This we can define from the negation of distance T ∗ ,
where T ∗ is the dual t-conorm of the t-norm T . A mathematical explanation is
given in [16], without entering into details, the distance between the attributes
of the agents makes a comparison of how ”far (different) are these attributes”
so that, the negation of this gives us how they are similar.
If we with α mark the attributes of the agent i, and with A mark the set
with these attributes, then Rsimilarity will be calculated as:

Rsimilarity (Ind, Ind2) = IF OW A(∀αi ∈ {Ind, N (d(αi (Ind), αi (Ind2))})


Where IFOWA is an aggregation of multi-criteria procedures. After spec-
ifying how the ordered weights w (whose sum is always 1), it is possible to
change the form of aggregation. Through these weights it is possible to control
the importance of each attribute in the global similarity. Let’s focus on direct
interaction between agents. As mentioned earlier in this complex adaptive sys-
tem, there are several attributes that affect one another. The attributes, such as
gender or age, that characterize the participants in the crowd, in which they can
not be influenced by other attributes, but some, like ideology, economic status,
status in society, are influenced by political trends or other attributes. This local
influence on a person with the definition of ”fuzzy concepts” can not be easily
established. After a detailed analysis, as well as with the help of social science
experts and using the models that were used previously, we can give following
definition:
Definition 1 Let A be an IFS in U that expresses attributes (characteristic) of
a man. Let ∆A is the variation (changes) of the attributes A of the corresponding
agent Ind with its environment ∆A : U → [0, 1], which determines the impact of
characteristic A from the average of each agent Ind and can be defined as the
aggregation of all influences. This influence is determined by the ”proximity” of
the agent ”Ind”, the distance d between the selected attributes is expressed by:
∆A (Ind) = IF OW Ai=1,...,N (Rsimilarity (Ind, Indi ) AN D d(X(Ind) − X(Indi )))
8 Sh. Ismaili, S. Fidanova

We will use the IF t-norm product to calculate the AN D conjunction. Let


Rsimilarity (Ind, Ind2) : U × U → [0, 1] is IF in a set of individuals who give
them a degree of ”similarity”. This IF relation is determined by the aggregation
(OWA) classical relation ”fqinje” with the fuzzy relations Ragent and Rpolice .

Rsimilarity (Ind, Ind2) =


= IF OW A(Rf qinje (Ind, Ind2), Ragent (Ind, Ind2), Rpolice (Ind, Ind2)) =
= w1 ∗ Rcouple (Ind, Ind2) ⊕ w2 ∗ Ragent (Ind, Ind2) ⊕ w3 ∗ Rpolice (Ind, Ind2)
and the evolution of one attribute is determined by each individual as well

A(Ind) = OW A(A(Ind), ∆A (Ind))


Another important aspect of agents is their condition. The agent’s condition
is defined by his position in society and determines his behavior. Therefore, an
agent in a ”peaceful” state who is an accomplice in power can not influence,
while an agent with no suitable position will be more likely to join the crowd.
But where are the boundaries between the states? In IF systems this will depend
on attributes: greed, legitimacy, and risk to be closed. Because there is not such
a strict limit here, it is necessary to apply the IF logic.

4 Conclusion
In this paper is represented a model of interaction between agents in multi agent
system in a case of protest. Different events and level of interactions are simulated
using IF sets and IF relations. In complex social systems some of the variables
are not strongly determined. Some of the values can have level of uncertainty.
Therefore IFS gives a possibility to represent social systems in more realistic
way.

Acknowledgment
Work presented here is partially supported by the Bulgarian National Scientific
Fund under the grants DN02/10 “New Instruments for Knowledge Discovery
from Data, and their Modeling” and ”Efficient Stochastic Methods and Algo-
rithms for Large-Scale Problems”.

References
1. Atanassov K., Intuitionistic fuzzy logics, Springer, 2017.
2. Atanassov K., On intuitionistic fuzzy sets theory, Springer, 2014.
3. Atanassov K.T., Intuitionistic fuzzy sets: theory and application, Springer, 1999.
4. Chellapilla, K., Fogel, D.B., Evolving neural networks to play checkers without
relying on expert knowledge, IEEE Transactions on Neural Networks, Vol. 10(6),
1999. 1382 - 1391.
Title Suppressed Due to Excessive Length 9

5. Collier, P., Berdal, M. and Malone, D. M. (eds.), Greed and Grievance: Economic
Agendas in CivilWars, Lynne Rienner Publishers, 2000.
6. Davies, T. P., Fry, H. M., Wilson, A. G., Bishop, S. R., A mathematical model of
the London riots and their policing, Scientific Reports, Vol. 3, article 1303, 2013.
7. Deshrijev G., Cornelis C., Kerre E., On the representation of Intuitionistic Fuzzy
t-Norms and t-Conorms, IEEE Transaction on fuzzy system Vol. 12(1), 2004, 45-61.
8. Epstein, J. M., Generative social science:Studies in Agent-Based Computational
Modeling, Princeton University Press, 2007.
9. Epstein, J. M., Modeling civil violence: An agent-based computational approach,
Proceedings of the National Academy of Sciences of the United States of America,
vol. 99, 2002, 72437250.
10. Gilbert,. N., and Troitzch, K., Simulation for the social scientist, Open University
Press, ISBN-13 978 0335 21600 0, 2005.
11. Ilachinsky, A., Artificial War. Multiagent-Based Simulation of Combat, World Sci-
entific, 2004.
12. Jager, W., R. Popping, R., van de Sande, H., Clustering and Fighting in Two-
party Crowds: Simulating the Approach-avoidance Conflict, Journal of Artificial
Societies and Social Simulation, Vol. 4(3), 2001, 19p.
13. Lomborg, B., Nucleus and Shield: The Evolution of Social Structure in the Iterated
Prisoners Dilemma, American Sociological Review 61, 1996, 278307.
14. Ejegwa P.A., Akowe S.O., Otene P.M., Ikyule J.M., An overview on intuitionistic
fuzzy sets, International Journal of Scientific & Technology Research, Vol. 3(3),
2014, 142-144.
15. Situngkir, H., On massive conflict: Macro-micro link, Journal of Social Complex-
ity,Vol. 1(4), 2004, 1-12.
16. Valverde, L., On the structure of f-indistinguishability operators, J. Fuzzy Sets and
Systems, Vol. 17(3), 1984, 3313-328.
17. West, J.E., Linster, B. The Evolution of Fuzzy Rules as Strategies in Two-Player
Games, Southern Economic Journal 69, 2003, 705717.
18. Xu Z.S.: Intuitionistic fuzzy aggregation operators. IEEE Transactions on Fuzzy
Systems, Vol. 15(6), 2007, 1179-1187.
19. Zadeh L.A.,Similarity relations and fuzzy orderings, Information Science Vol. 3(2),
1971, 177-200.

You might also like