Role of Trust in Creating Opinions in Social Networks
Jiří Jelínek
Institute of Applied Informatics, Faculty of Science, University of South Bohemia,
Branišovská 1760, 370 01, České Budějovice, Czech Republic
Keywords: Trust, Social Systems, Opinion Dynamics, Multi-Agent Systems.
Abstract: Although we are not always aware of this, our existence and especially communication are based on the
principles of trust. The importance of trust is crucial in systems where risk is present e.g. when handling the
information we have acquired through communication because it is not always possible to immediately
verify the truthfulness of it. The aim of this paper is to link two areas, namely the reality in the human
community described above and the available knowledge of social networks and multi-agent systems, and
try to simulate real trust concerned scenarios in society by these tools. The multi-agent model will be
presented, which simulates the behavior of the heterogeneous group (people-like) entities in the process of
creating their opinion about the world on the basis of information acquired through communication with
other agents. The focus is placed on the processes influencing trust in communication partners and its
dynamics. The results of the experiments are also presented.
1 INTRODUCTION
Although we are not always aware of it, our
existence and communication in particular are based
on principles based on moral values in society. The
key role here is played by the trust between
communicating parties, the importance of which is
crucial to further handle the sometimes unverified
information we have acquired through these
communications.
Today we often hear statements made by people
in the media that are either unsubstantiated or
frankly false. Some individual from society could
react to these statements, making it clear that their
opinion is somewhat different. However, these false
statements frequently go unnoticed in the media, and
then it is up to the individual to state the conformity
or inconsistency of the statement with his/her
opinion of the world. In case of an inconsistency,
confidence in the given source of information is
reduced.
The consequences of both cases mentioned are
very serious and lead to increased caution when
communicating and place an emphasis on an
individual’s own knowledge. These individuals also
explore the trustworthiness of the communication
partners and are able to better consider whether the
communication brought usable information.
We could consider this topic as a purely
philosophical one, but it is not. The same problems
can be observed in the information systems and
multi-agent models and also in the Internet of
Things. The time has come when well-known people
begin to call for a basic codification of the IoT
environment in terms of ethics and accountability
(Cerf, 2017).
The aim of this paper is to link these two areas,
namely the reality in the human community
described above and the available knowledge of
social networks and multi-agent systems, and to try
to simulate real trust scenarios by these tools.
The next sections of the paper are organized as
follows: section 2 focuses on the current state of the
knowledge and a discussion on the selection of
literature dealing with this field. Section 3 describes
the model presented and its configuration and
classification. Section 4 then focuses on
implementing the model and experiments.
2 STATE OF THE ART
The problem of trust has gone through several
periods in the past when it was more emphasized.
The last two of these periods took place around
2002-2006 and during 2012-2016. A large number
208
Jelínek, J.
Role of Trust in Creating Opinions in Social Networks.
DOI: 10.5220/0006592102080215
In Proceedings of the 10th International Conference on Agents and Artificial Intelligence (ICAART 2018) - Volume 1, pages 208-215
ISBN: 978-989-758-275-2
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
of contributions mainly focused on the field of
multi-agent systems can be found during these
periods and beyond, which were concerned with the
question of trustworthiness in communication.
Surveys such as (Ramchurn et al., 2004), (Pinyol
and Sabater-Mir, 2013) and (Granatyr et al., 2015)
are especially important for examining the issue so
as the introductory commentary for the conference
section (Falcone and Singh, 2013). The
classification of existing models of trust is also
proposed in (Granatyr et al., 2015). In section 3.4 we
classify our model according to these criteria. Two
approaches to working with trust in a partner are
presented in (Ramchurn et al., 2004) - individual and
systematic one. We use the elements of both of
them. Our model is based on an individual approach
and the credibility of the environment is guaranteed
on a system-wide level.
A well-known but often difficult to describe
relationship of trust and risk is mentioned in the
paper (Varadharajan, 2009). Also the validity of the
partner’s trustworthiness is presented as dependent
on the type of service, message content, or field of
expertise. However, the frequent service overlap and
link to an individual rather than a service is cited in
other materials. We think that trust can be shared in
areas similar to one another, where one can expect
a similar competence of a partner. In our model, we
suppose trust is connected to an individual.
The next paper (Yolum and Singh, 2003) focuses
on a specific area of service provision. The model
described focuses primarily on the system of
transmitting references between agents about service
providers. The quality of these services is evaluated
by agents, but the model does not work with
verifying the messages. We considered this verifying
mechanism to be important and thus we
implemented it in our model.
Regarding the delivery of the service, ideas are
presented in the paper (Sen, 2013). The paper
presents some interesting notes about trust in
general, such as the relationship of trust and the cost
of the service or the communication. This idea is
interesting and we would like to implement it in our
model.
The problem of anonymity is discussed in
(Fredheim et al., 2015), which presents the study of
the behavioral changes on a discussion forum before
and after the mandatory identification. This area is
generally neglected in multi-agent systems. Our
model is not an exception, but anonymity will be the
subject of further development of it. The current
status can be described as identification with the
pseudonym.
On the contrary, the article (Huang and Fox,
2006) coincides partially with our approach in
focusing on the dissemination of information in
social networks. The issue of possible transitivity of
trust is also raised. However, from our point of view,
it is more appropriate to set the agent’s trust in their
partner’s partner as a product of the agent’s trust in
their partner and the partner’s trust in their partner.
This respects the experience and trust in partners.
A model examining the spreading of information
in the social network can be found in (Jelínek,
2014). The experience gained when constructing this
model was used in this contribution, especially the
rating mechanism of communication partners,
defined on the basis of individual communication
between agents.
The field of monitoring the developments on the
dynamics of individual opinions in conjunction with
political or similar influences is focused in (Horio
and Shedd, 2016). The authors present their work in
the field and the main published models - Deffuant-
Weisbuch (DW) (Deffuant et al., 2000) and
Hegselmann-Krause (HK) (Hegselmann and Krause,
2002). For our model, the statement based on
(Lorenz, 2007) is interesting - when forming
opinions (communication for this purpose),
individuals prefer partners with similar opinions.
3 DESCRIPTION OF THE
MODEL
The area targeted by this contribution is specific due
to the focus on disseminating information inside the
community and creating a world opinion, especially
in relation to mass communication. We focus on
situations where verification of truthfulness of
transmitted messages is possible only with a delay
and sometimes not possible at all.
The specificity of the presented model is also in
using an information (knowledge) base of individual
agents, which can be presented as a basis on which
the opinion of the individual about the world is
formed. We suppose expression of trust with a single
number in the range (0, 1), where 1 means absolute
trust. We also distinguish between two types of trust
- the trust in the partner and the trust in the content
of the information.
3.1 Model Principles
Three logical levels can be identified in the model.
At the highest level, the model focuses on the
Role of Trust in Creating Opinions in Social Networks
209
trustworthiness of communication in the social
network while examining the problems of the truth
of the information transmitted. The subsequent
influence on the development of opinions of the
individuals in the community is also taken into
account. This topic is related to the issue of trust at
a higher level - trust in general. This is also where
the influence of information verification is
examined.
The middle level is agent’s trust setting in the
multi-agent environment itself. Here, the model uses
the local tools - agents independently evaluate the
credibility of their partners with the help of the
content of the message. The main processes of
communication and setting and working with trust in
partners are defined at this level. The model is
capable of implementing agents with different
settings and behaviors (a heterogeneous
community).
The third level of the model is the representation
of transmitted information and the specific methods
used for calculating the necessary parameters and
their relation to higher levels of the model. We use
a representation of knowledge with the N3 clauses
(subject – link - object) complemented by
a continuous sureness parameter within the interval
(-1, 1). The sureness is an inseparable part of N3
knowledge representation, so we talk about the
extended N3 knowledge representation. The clause
is further supplemented with metadata about the
source (from whom it was obtained), the origin (who
originally created it), the trust in it, its verification
state and its activity state (related to the process of
forgetting). The last three metadata are continuous
values on the interval (0, 1).
Exactly one clause is transferred during one
communication between agents, and each agent
gradually creates its own knowledge base
represented by a set of these clauses. The base can
be represented as an oriented graph with subjects
and objects in the nodes and links between them.
The graph is assumed to preserve the transitivity of
the links for calculating some of the values,
(if
The agent’s trust in the partner (i.e. the hope the
partner is able to provide us with the information
and also wants to do so) is set purely individually
within the inner processes of the agent. The agent is
limited to its own experience when setting the trust.
3.1.1 Information Verification
The truthfulness of the message the recipient may or
may not be able to be verified. The time factor is
also important (verification may not take place
immediately when receiving the message but later).
The key factor is whether the recipient has access to
the resources that allow the information to be
verified.
We could also discuss what source the agent
would consider being sufficiently objective (trusted)
to verify the message. However, this creates
a vicious circle because the same aspects are
relevant for this source as are for the sender of the
message. One possible way is to define an authority
that will objectively store truthful information about
the world and will be able to verify the message and
truthfully inform the questioner. This authority in
the model is a special agent called the world. The
information (clauses) provided by the world are
generally correct and verified but agents do not have
to know that immediately. We call the
communication of the agent with the world and
obtaining information from it as an observation.
3.2 Selected Model Details
Due to the limited space of this contribution, some
key functionality has been selected from the model,
which will be further explored in the following
subsections.
3.2.1 Clause Usefulness
The model works with the knowledge base of the
agent, over which it quantifies several variables. The
first of the key parameters is the usefulness u
c
of the
given clause from the interval (0, 1).
The clause is useful if it delivers information the
agent does not have on his or her base and cannot be
derived from it, or has it, but with less sureness or
worse metadata. Deriving, in this case, means to find
the shortest path in the base graph starting with the
subject of the clause and ending in the object where
the type of all the links is the same as the link in the
clause.
If the new N3 clause has a higher sureness
(positive or negative), it is useful for the agent and
included in the base.
If this is not the case, the path described above is
searched for, and if it exists, the metadata of the
clauses contained in the path are investigated. The
aim here is determining the value of the metadata for
the entire sequence of clauses. According to the
principle of the weakest part of the chain, it will be
the lowest value of the whole path, so the minimum
value of verification v
min
and trust t
min
is sought. If
v
c
> v
min
and therefore the new clause is better
ICAART 2018 - 10th International Conference on Agents and Artificial Intelligence
210
verified, it is useful in u
c
= v
c
- v
min
. If this does not
apply, the confidence is calculated in the same way,
i.e. for t
c
> t
min
the usefulness of clause is u
c
= t
c
-
t
min
.
If the path does not exist in the base, the benefit
is set according to the N3 clause element
information. Usefulness is set as
(1)
where f
x
are values calculated for three elements of
N3 (subject, link, object) as
(2)
The d
x
value is the degree of the given node or the
frequency of the given link type in the base. This
calculation encourages the agent's behavior to
expand its knowledge base and, therefore, clauses
containing N3 elements not yet included in the base
or included with a low degree or low frequency are
more useful.
3.2.2 Trust in Clause
Trust in the clause is generally based on its content,
but our confidence in the source from which we
obtained it is also significant. The weights of these
parts are matters of the personality profile of the
individual and therefore the model must allow them
to be modified. The specific confidence calculation
method in the t
c
clause is then set according to the
formula
(3)
where k
s
is the personal self-trust factor of the
individual in his own knowledge in the range (0, 1),
s
c
is the sureness of the N3 clause, s
b
then the
sureness of the N3 derived from the individual's
knowledge base as the minimum value of the
sureness of the same N3 clauses in the base or the
path found by derivation (see 3.2.1.). The
t
s
parameter is the agent’s trust in the sender and
t
p
the sender’s trust in that clause. Obviously, for
k
s
= 1 the individuals will rely solely on their
knowledge and vice versa.
3.2.3 Trust in Sender
The value of t
s
is the subject of a further description
of the model. In principle, this trust in the source of
information will certainly be based on verifying the
information obtained from it. However, it is also
necessary to consider a state where verification is
not available at a given moment or is not available at
all. The t
s
value evolves over time and describes the
long-term experience with the partner. This
dynamics is expressed by a classical mechanism of
the gradual modification of the value t
s
according to
the formula
(4)
where k
t
is the adjustment factor with value in the
interval (0, 1). The value x
c
in the formula is either
the value v
c
(if verification is available) or the trust
value t
c
in other cases.
3.3 Model Parameters and
Initialization
The model is designed with high flexibility and the
user can set it using the five parameters for the entire
model:
• The number of individual objects occurring as
objects or subjects in clauses.
• The number of link types occurring in clauses.
• The number of clauses forming the knowledge
base about the world administered by the world
agent.
• The initial number of agents in the simulation
(which can change).
• The number of simulation steps.
When selecting the agent parameters, the aim was to
eliminate some of the limits of the existing models
identified in section 2. E.g. it means implementing
the self-trust factor (see 3.2.2) and the probability of
choosing a random partner (reflecting the random
communication in the community). The next seven
parameters determine the agent’s behavior:
• The probability of observation, i.e. acceptance of
a new clause from the world agent.
• The probability of accepting a new clause from a
partner in communication.
• Agent forgetting factor affects (decreases) the
activity of the given clause or orders its deletion
when the activity is very low.
• The probability of choosing a random partner
which takes into account random
communications.
• The probability of selecting a random clause in
communication. Otherwise, it is preferred to
choose a partner’s clause that contains the same
N3 elements as the one obtained last.
• The rate of increasing trust. It is the speed at
which trust in the partner providing useful
information grows.
• Self-trust rate. The higher, the more the agents
rely on themselves and their knowledge.
Role of Trust in Creating Opinions in Social Networks
211
It is clear that model setting is a multidimensional
problem and the model is able to simulate various
scenarios according to specific requirements. On the
other hand, the number of degrees of freedom
represented by the number of model parameters
complicates validating the model.
Already during the first experiments with the
model, it turned out that the model is very sensitive
to the initial settings, especially in terms of whom
the agents prefer to communicate with. The model
provides three initialization options.
The first option is to choose a scale-free model
with random communication partners (random
initialization - RI). The agent has complete freedom
to choose a partner.
The second option is orientation on the preferred
sources of information (preferred initialization - PI).
Network dynamics should respond faster to world
changes.
The third option is the second approach
alternative but now preferred the locality (local
initialization - LI). Agents use the closest simple
agents (in terms of their numbering).
3.4 Model Classification
Classification of the model was performed in order
to fit it into the overall issue of research on
trustworthiness. The criteria presented in (Granatyr
et al., 2015) were used in the following text in the
form of a dimension-value-description.
Paradigm type - cognitive, numerical. The model
imitates the behavior of human individuals but also
takes into account the numerical procedures based
on the processing of historical data and the content
and structure of the information transmitted.
Information sources - direct interaction (DI),
partly witness information (WI), partially certified
reputation (CR). Data collection from direct
interaction between agents (DI) is the key to
establishing trust but the information about trust or
origin transmitted from previous sources and
recorded on the transmitted message are also used
(WI). In the part of the model (agent world) the
general validity and full trust (CR) is assumed.
Cheating assumptions - cheating (L2). The
model has no assumptions in the area of false
information and cheating is allowed. However, the
mechanism for verifying a message that is able to
reveal the liars is used, even if this does not
necessarily become immediately.
Trust semantics - partially. Trust is represented
in the model as the only number but it covers several
areas with different weights and clear semantic
significance.
Trust preferences - partially. The model works
with the weights of components from which trust is
calculated. These weights characterize the
personality of the agent and his preferences and are
given as agent parameters. The agent can change
them (but this is not being used now).
Delegation trust - no. The delegation concept is
not used in our model.
Risk measure - partially. The risk of choosing
a trustworthy communication partner can be
determined but not necessarily. The key here is if it
is possible to verify the quality of the selected risk
value by verifying the content of the message.
Incentive feedback - no. Our model assumes that
agents truthfully inform on the metadata of messages
and communications.
Initial trust - no. Our model does not provide
a special initial trust setting for newcomers, but it
does not penalize them. The initial trust of a new
agent is set to a neutral value 0.5.
Open environment - partly. The model is open
and it does not address possible fraud (change of
agent identity).
Hard security - no. The possibility of a security
breach in communication is not assumed.
4 IMPLEMENTATION AND
EXPERIMENTS
The presented model was implemented to verify its
ability to simulate the dynamics of communication
and creating individual knowledge bases in the
community of individuals. Each agent
communicates with the others on the basis of
a random formula. On the basis of this
communication, individuals formulate the opinion
about the world in which they live.
The model has been implemented in Java and
one of its undisputed benefits is very detailed logs of
all significant parameters of each agent and the
overall model (millions of values for experiments
presented).
The description of the world is generated by the
world agent. Its knowledge base characterizes the
objective state of the world without any distortion
caused by the sensors or due to the trust setting. The
clauses can then be disseminated to other agents
through the communication.
ICAART 2018 - 10th International Conference on Agents and Artificial Intelligence
212
4.1 Experiments
The purpose of the experiments was to simulate and
then to analyze the behavior of individuals in an
environment with the limited possibility of verifying
the information or knowledge. The performed
experiments focus on specific scenarios formulated
with the aim to match the real world situations very
closely. Only the selection of results demonstrating
a given tasks has been included in this article.
The basic setting of experiments was an
environment in which only a limited group of
privileged agents has access (two in our
experiments). These agents gain objective
information by observing the world. One of them
does this without a change of information (a good
agent), but the other one intentionally manipulates
the information and negates the sureness of clauses
(a bad agent).
In addition to these privileged agents, there is
a set of other simple agents without access to the
world. They create their knowledge bases only on
the basis of communication with other agents,
including the privileged ones. With the above-
mentioned bad agent exception, it is assumed that
this communication takes place without distortion
and a change of the information. The factors studied
here are the knowledge bases of individuals and
their development. The underlying hypothesis here
is that an individual can develop his world opinion
on the basis of completely erroneous information of
a bad agent.
The similarity of the knowledge base of the agent
with the knowledge bases of privileged agents was
chosen as an output value. This is determined by
comparing each simple agent’s knowledge clause
with a privileged agent base. If the given clause (its
N3 parts) can be derived from the base, the clause is
considered to be similar according to the difference
of sureness between the clause and the base. The
resulting summary across all of the simple agent’s
clauses is then normalized by the size of the agent’s
knowledge base. It does not apply that the sum of
similarities to all privileged agents is 1 - the simple
agent more similar to a good one can be partially
similar also to bad one.
The second output is the trust (and thus
preference) in agents in the social network. It is
individually calculated by a particular agent, but its
average value can be determined by the formula
(5)
where N
j
is the number of agents having among their
partners agent j and d
ij
is the trust of agent i in agent
j.
The global model setting was the same for all
experiments. There were 20 individual objects and
5 types of links from which 100 clauses were
randomly generated. This was sufficient for the
selected scenarios, but it would be interesting to
examine the influence of the world knowledge base
size on the model behavior. The number of simple
agents was set to 50; the number of simulation steps
(x-axis of all graphs) was usually 500. The
experiments also used different settings for
individuals (see scenarios).
4.1.1 Scenario 1
The first scenario was a simulation of network
dynamics without the possibility of verifying clauses
and the process of forgetting. Its goal was to
investigate how the initialization method affects the
behavior of agents.
Figure 1: Average similarity of agents to and trust in good
or bad one – RI.
Three experiments were performed based on the
initialization type. The first was random
initialization (RI). Its output normalized by the total
number of agents is shown in Figure 1. Because the
simple agents can also communicate with each
other, we can see at the beginning about 100 steps
period, when simple agents create the base without
clear orientation to good or bad agent knowledge
(the trust in these agents is still evolving). The
situation stabilizes after this period of profiling.
It is clear that trust in both privileged agents is
essentially the same, as well as the number of
clauses from them in the knowledge bases of
ordinary agents. However, every simple agent
gradually developed into a supporter of one of the
privileged (what can be seen from other outputs).
Role of Trust in Creating Opinions in Social Networks
213
The same scenario for initialization with the
preference of direct communication with privileged
resources (PI) is shown in Figure 2. We see that trust
in privileged agents has slightly increased. However,
the greatest change is evident at the beginning of the
simulation when agent profiling went very quickly.
For this experiment, we can also observe (again
from another model outputs) the privileged agents'
preference in communication and also the highest
values of trust in them from all agents.
A situation very similar to a random initialization
occurs in the case of initialization with the closest
partner preference (LI). Agent profiling again caused
significant fluctuations at the beginning of the
simulation.
Figure 2: Average similarity of agents to and trust in good
or bad one – PI.
4.1.2 Scenario 2
The objective of the second scenario was to verify
how dynamics of the network changes when
allowing the clauses to be verified. The verification
was enabled from step 200, the other settings were
the same as in the first experiment from the previous
scenario, except the number of simulation steps
(here 1000 steps).
Figure 3 shows a change in the behavior of the
model and the increase of the average similarity of
the agents' bases with a good agent. The agents
reoriented themselves to resource that offered
information consistent with the real world. However,
the opinion of agents originally oriented on a bad
agent is changing very slowly. Also, the trust in
a bad agent has not fallen to the simulation end but
has grown significantly slower than trusting in
a good agent.
Figure 3: Average similarity of agents to and trust in good
or bad one – verification enabled.
The position of the bad agent appears to be quite
good but, on other outputs, it can be shown that this
is not the case. In a purely binary classification of
similarity, the orientation on a good agent is the
dominant one from the step about 350.
In Figure 3 (on the right-hand y-axis in percent),
is dottted the average size of agents’ bases.
4.1.3 Scenario 3
The last simulated scenario was focused on the
effect of forgetting on network dynamics. The initial
settings were taken from Scenario 2 and modified by
enabling forgetting of clauses in agents’ bases. The
number of simulation steps was also reduced to 500
steps.
Figure 4: Average similarity of agents to and trust in good
or bad one – verification and forgetting enabled.
In Figure 4 the meanings of the lines are the
same as in Figure 3. By comparing both Figures, it is
obvious that enabling the forgetting of older clauses
has greatly accelerated the process of agent
orientation to the correct source (good agent). Yet in
ICAART 2018 - 10th International Conference on Agents and Artificial Intelligence
214
addition to that, agents’ knowledge bases were
reduced in size on average of about 2.5 times
(measured in step 500). This means that agents have
much less knowledge for creating opinions, but the
knowledge is of higher quality (clear profiling on the
good agent). The smaller knowledge base also
significantly contributed to the speed and efficiency
of agents’ activity.
5 CONCLUSIONS
This paper presents a multi-agent model simulating
the behavior of the heterogeneous group of (people-
like) entities in the process of creating their opinion
about the world on the basis of information acquired
through communication with other agents. The
reason for the constructing the model is to
investigate the dynamics of trust in an environment
with limited possibilities to verify the transmitted
information. Local metadata from previous contacts
with the partner are used for establishing trust as
well as the knowledge base of the agent. Presenting
techniques to evaluate this information and use it in
trust settings were presented as well. The model is
highly adjustable with global and agent-specific
parameters.
The performed experiments were focused on
scenarios where the limited possibilities to verify the
information caused that the agent's knowledge base
could be built on completely incorrect information.
The results also showed that the later availability of
verified information changes the knowledge base
and hence the attitudes of the individuals very
slowly. It was also shown that forgetting older
clauses from the knowledge base leads to quicker
trust profiling of simple agents, and to accepting
knowledge primarily from the verified source.
The created model will be further developed and
tested especially on the basis of ideas from papers
cited in section 2. Future work should be
concentrated on analyzing the problem of dynamics
of the community structure. Special attention will be
laid on model validation, where it will be necessary
(due to specific model parameters) to collect real-
world data for deeper model validation.
This model could be increasingly used in the
future, depending on how individuals and companies
gradually discover the possibility of manipulating
information. Typical examples of applications can
be social systems where individuals can spread
unverified or false information and systems with
limited ability to verify information.
REFERENCES
Cerf, V. (2017). A brittle and fragile future.
Communications of the ACM, 60(7), pp.7-7.
Deffuant, G., Neau, D., Amblard, F., & Weisbuch, G.
(2000). Mixing beliefs among interacting
agents. Advances in Complex Systems, 3(01n04), pp.
87-98.
Falcone, R., & Singh, M. P. (2013). Introduction to special
section on trust in multiagent systems. ACM Trans. on
Intelligent Systems and Technology (TIST), 4(2), p. 23.
Fredheim, R., Moore, A., & Naughton, J. (2015).
Anonymity and Online Commenting: The Broken
Windows Effect and the End of Drive-by
Commenting. Proceedings of the ACM Web Science
Conference (p. 11). ACM.
Granatyr, J., Botelho, V., Lessing, O. R., Scalabrin, E. E.,
Barthès, J. P., & Enembreck, F. (2015). Trust and
reputation models for multiagent systems. ACM
Computing Surveys (CSUR), 48(2), 27.
Hegselmann, R., & Krause, U. (2002). Opinion dynamics
and bounded confidence models, analysis, and
simulation. Journal of artificial societies and social
simulation, 5(3).
Horio, B. M., & Shedd, J. R. (2016). Agent-based
exploration of the political influence of community
leaders on population opinion dynamics.
In Proceedings of the 2016 Winter Simulation
Conference (pp. 3488-3499). IEEE Press.
Huang, J., & Fox, M. S. (2006). An ontology of trust:
formal semantics and transitivity. In Proceedings of
the 8th int. conference on Electronic commerce (pp.
259-270). ACM.
Jelínek, J. (2014). Information Dissemination in Social
Networks. In Proceedings of the 6th Int. Conference
on Agents and Artificial Intelligence-Volume 2 (pp.
267-271). SCITEPRESS.
Lorenz, J. (2007). Continuous opinion dynamics under
bounded confidence: A survey. International Journal
of Modern Physics C, 18(12), pp. 1819-1838.
Pinyol, I., & Sabater-Mir, J. (2013). Computational trust
and reputation models for open multi-agent systems: a
review. Artificial Intelligence Review, 40(1), pp. 1-25.
Ramchurn, S. D., Huynh, D., & Jennings, N. R. (2004).
Trust in multi-agent systems. The Knowledge
Engineering Review, 19(1), 1-25.
Sen, S. (2013). A comprehensive approach to trust
management. In Proceedings of the 2013 int.
conference on Autonomous agents and multi-agent
systems (pp. 797-800). International Foundation for
Autonomous Agents and Multiagent Systems.
Varadharajan, V. (2009). Evolution and challenges in trust
and security in information system infrastructures.
In Proceedings of the 2nd int. conference on Security
of information and networks (pp. 1-2). ACM.
Yolum, P., & Singh, M. P. (2003). Ladders of success: an
empirical approach to trust. In Proceedings of the
second int. joint conference on Autonomous agents
and multiagent systems (pp. 1168-1169). ACM.
Role of Trust in Creating Opinions in Social Networks
215
