Information Dissemination in Social Networks
Jiří Jelínek
Institute of Technology and Business in České Budějovice, Okružní 517/10, České Budějovice, Czech Republic
Keywords: Social Network Dynamics, Agent based Simulation, Information Dissemination.
Abstract: Social networks are currently one of the most studied structures for information and knowledge exchange.
These structures are very well described in terms of their static structure, this article attempts to propose the
model of their dynamic behavior and the spread of information and knowledge in these networks. The
heuristic event based model of the individual behavior in the network using the message passing will be
presented. The main idea of the model is agent’s need for information and knowledge in specific situations.
On that is based multiagent model of the social network used for information exchange, which has been
practically implemented and will be presented as well as some of its simulation outputs for tasks testing the
dynamics of social networks.
1 INTRODUCTION
Social networks are currently the most studied
structures for information and knowledge exchange.
These structures are very well described in terms of
their static behavior; our contribution tries to
describe the dynamics of information dissemination
within them.
Social network can be thought of as a graphical
structure whose nodes are objects or individuals
(people) and edges represent some kind of
connection between these objects. The nature of
these connections may vary but they always express
a certain link or communication between connected
nodes. During the time the structure of the network
varies; usually as a result of the communication
between nodes (individuals).
This paper aims to show one possible way of
modeling the dynamics of information dissemination
in a social network. The first experimental results
show some phenomena that were seen in the
structural and information dynamics of modeled
networks.
The rest of the article is structured as follows. In
chapter 2 the actual state in simulation of social
networks dynamics is presented. Chapter 3 is
focused on principles and implementation of the
proposed model. In chapter 4 are presented
preliminary experimental results. Some conclusions
and recommendations for future work are
summarized at the end of the article.
2 STATE OF THE ART
For social networks modeling is very frequently
used the agent based (AB) approach, which is
preferred where local data about individual’s
behavior are available or where global probabilistic
data allowing to set individual parameters can be
inferred. The popularity of AB approach increases
the quantity and quality of modeling tools, further
information can be found in (Siebers at al., 2007).
For single agent modeling we can use any technique
which allows describing the agent behavior (e.g.
system dynamics, event based modeling, state
charts, flowcharts).
The social networking topics are discussed in
a number of publications, an overview of social
network analysis techniques can be found in
(Wasserman et al., 1994). Several social metrics of
the network useful for decision making including the
FRINGE algorithm are also presented in (Palazuelos
et al., 2012).
The rating of agents which influences the links in
the network and its stability is described in (Wu et
al., 2011). The idea of agent rating is also used in
our approach.
Many papers are focused on analysis of the real
data. Very interesting analysis of large real network
dynamics can be found in (Zhao et al., 2012). The
use of agent based approach in the analysis of social
networks (e-mail based) is addressed in (Menges et
al., 2008). Very actual is the field of social networks
267
Jelinek J..
Information Dissemination in Social Networks.
DOI: 10.5220/0004917202670271
In Proceedings of the 6th International Conference on Agents and Artificial Intelligence (ICAART-2014), pages 267-271
ISBN: 978-989-758-016-1
Copyright
c
 2014 SCITEPRESS (Science and Technology Publications, Lda.)

in Internet-based games. E.g. the analysis of the
social network in bridge community is presented in
(Balint et al., 2011).
The spread of ideas in online networks is also
engaged in Ahmad’s work (Ahmad et al., 2006),
where authors use several measurements, such as
acceptability, portability and accessibility. Our
model was also affected with this approach. Visual
analytics approach for analysis of network dynamics
is described in (Federico et al., 2011). The work of
Ma (Ma et al., 2013) is focused on detecting
communities and their tracking in the dynamic
network; new algorithm is proposed for this goal.
The idea of the event based social network model
is presented in (Jelínek, 2011). The model is based
on the agent’s need for information to find the best
possible reaction on randomly occurring events. The
event is here a very general concept and we can
imagine different thing under it, e.g. some situations
in the environment, messages sent to agent, etc. Our
model uses the same basic principle.
3 MODEL OF THE SOCIAL
NETWORK
The model uses the multi agent approach based on
the coexistence and interaction of elementary objects
(agents). The state charts with event activated state
transitions are used at the agent level to describe his
behavior.
As the base of the model we assume the
existence of the agents who are exposed to some
"life" situations requiring their reaction (solution of
the situation). Agents want to make the best
solutions. They use their own information and skills
but also try to find the solutions of the same
situation created by the other agents in the network
in the past. The model is based on the closed world
assumption applied to the number of nodes in the
network and the set of possible situations which is
constant and unchanging over time. This assumption
is an acceptable simplification of the reality. The
dissemination of knowledge for solving new
situations can be simulated with correct setting of
the time when situations will arise. Our goal is to
examine how agents improve their skills (the ability
to use the best reaction on the situation).
The detailed modeling of the “life” situations is
not a crucial issue; they are represented just like
a message. The "solution" to every possible situation
j from the total number of N is represented as
a single number s
j
from the interval <0, 1> which
also describes the quality of reaction (1 = highest
quality, that means the best solution). In the future
versions of the model this number will be replaced
by function describing more precisely the quality of
reaction.
3.1 Detailed Description of the Model
The basic agent action in each simulation step is
described on figure (1)
.
Figure 1: Diagram of agent actions at each simulation step.
(Jelínek, 2011).
In each simulation step the environment sends
messages according their activation rules
(probability function or time based event) to
(usually) randomly selected agents.
The agent can generate a new solution based on
his education and knowledge level (simulated by the
agent ability to generate a solution with the quality
in the defined range). If the agent is part of a social
network, he also tries to find a solution through
asking his neighbors (connected agents - partners).
The different willingness of every agent to
communicate is respected so as the willingness to
accept the question and answer to it. These
parameters implement a simple model of agent’s
No
Start
Createlistofsolutions
frommemory
Addsolutionsfrom
friends(neighbors)
tothelist
Sugestnew
solution
Usebest
verifiedsolution
Usenotverified
solutionfromauthor
withbestrating
Storesolution
inmemory
Removeold
(notvalid)solutions
frommerory
Ifitispossible
verifysolutions
frommemory
Stop
No Yes
Someevent?
Yes
No
Yes
Anyverified
solution?
Anynotverified
solution?
ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence
268

personality and emotional state.
In the process of finding the best solution to the
given situation plays a crucial role the feedback, i.e.
the evaluation (verification) of the solution’s quality.
Without it, the agent is not able to prefer better
solutions. The model respects the fact that the
solution may not be verifiable immediately after its
adoption, but only after some period of time. The
information about solution verification is
represented by a special message sent to the agent.
The agent has its own memory for storing
previous reactions on different situations. It forms
the basis of a simple CBR (case based reasoning)
system. Every newly created solution agent stores
together with an identification of its author. The
author needs not necessarily be the agent from
whom the solution is obtained; it could be taken over
from another individual in the network. In the
memory is implemented the process of forgetting -
old solutions are removed from memory a certain
time after their inserting.
Information about the authors is used for rating
agents in the network. Author of the solution is
added to the list of partners and his rating is
modified in the moment of verification of proposed
solution. As we can see on figure (1), rating is used
in situations where there is not verified solution in
the agent’s memory or obtained from the network.
The rating is decreased when the partner does not
want to communicate and answer agent’s questions.
The length of the partners list (number of links) can
be limited and agents with lowest ratings are deleted.
The described mechanism is one of the ways how to
implement the principle of local trust describing the
oriented link between two agents.
Our main goal is to explore the dynamics of the
information dissemination in the network. To be
able to characterize the amount of knowledge of
each agent i, the quality q
i
according to the formula
(1) is calculated, where N is the total number
of situations and s
ij
the quality of the solution of
situation j used (generated or obtained) by agent i. If
a solution to the situation j is not available, the value
s
ij
is taken as zero. It’s taken into account that agent
may remember several solutions of specific
situation.



N
j
iji
s
N
q
1
)max(
1
(1)
The overall quality Q of the network was also
defined according to (2), where M is the total
number of agents in the network.



M
i
i
q
M
Q
1
1
(2)
Also the "popularity" of the agent as normalized sum
of agent’s ratings from all agents in the network was
defined according to (3), where p
i
is a measure of
agent’s i popularity and r
ji
is the agent’s i rating on
agent j (if agent i is not rated by agent j, the rating is
set to zero - r
ji
= 0).



M
j
jii
r
M
p
1
1
(3)
It is necessary to point out here that the rating is
conducted also on agent itself (it defines the opinion
of the agent about himself).
4 EXPERIMENTAL RESULTS
After the start of the simulation the network is set up
so that agents have links only with closest neighbors
(based on the distance between agents in 2D “world”
space).
Several parameters were watched during the
simulation experiments – especially the agent’s
quality q
i
and the overall network quality Q. Also
other data from the model can be monitored. Some
examples from the experiments are shown in the
following figures.
Figure 2: Network structure at the start of the simulation.
First experiments were performed on a network of
100 agents with the set of 10 situations. The
maximum number of agent’s links was 20. On figure
(2) is the network topology on the beginning (the
closest partners), on figure (3) the state after 500
simulation cycles. We can see the creation of
“information centers”, the agents with big popularity
(represented with the big radius of the circle) based
on their information competence and willingness to
communicate with others. The colour of the node
circle represents the quality of agent (darker circle =
better quality).
InformationDisseminationinSocialNetworks
269

Figure 3: Network structure after 500 simulation steps.
We can see that modeled social network may be split
into a number of non communicating agent groups.
This is in compliance with reality - closed
communities can get into the separation, which in
turn affects the knowledge quality of their members
because of limited access to information. To remove
this phenomenon, agents can use some "initiative" in
managing their lists of partners (e.g. using the model
of search services for finding new partners).
Figure 4: The overall network quality Q during the
simulation (simulation steps on x-axis)
Figure 5: Histogram of individual agent’s q after 500
simulation steps.
On figure (4) is presented the overall network
quality based on equation (2), and its progress
during the simulation. The y-axis shows percentage
of overall quality Q (100 is the maximum value,
every agent is able to use the best solutions for all
situations). Here we can see the dynamics of
information dissemination. On the beginning of the
experiment, the maximum number of agent’s
connections was set to 20; agents created their
personal networks (partner lists) and received the
knowledge about solutions of all possible situations.
In the step 300 of the simulation the maximum
number of connections was reduced to 2. This
reduction had minimal influence on the overall
network quality because the substantial connections
have already been created before.
We can also display the quality of individual
agents after the simulation. The corresponded
histogram is on figure (5).
We can compare the figures (4) and (5) with the
situation where the agents had no possibility to
communicate (number of links = 0 during the whole
simulation). The results are on figures (6) and (7).
Figure 6: The overall quality Q of agents during the
simulation (no network, simulation steps on x-axis)
From the results we can see the different behavior of
the agents and the big influence of the social
network on the quality of individual agents and the
whole “network” and also on the dynamics of the
information dissemination process.
Figure 7: Histogram of individual agent’s q after 500
simulation steps (no network).
ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence
270

5 CONCLUSIONS
Proposed model extends and improves the model for
simulating the dynamics of information
dissemination within the social network presented in
the earlier work of the author. The system of
messages is now used, which is crucial for future
work. The model can be used to study the dynamics
in social networks of varying size and orientation
created for the purpose of information exchange
(such as corporate networks, on-line services, local
structures aimed to solving everyday situations,
etc.), not limited to on-line services and electronic
transmission of information.
The presented results are only the beginning of
the exploring and simulating of the social networks
with this model, but experiments shown that model
provides interesting outputs comparable with the
behavior of individuals in the real world.
The model is continuously developing and
modifying. We plan to do much more experiments to
examine the influence of all parameters on the
network dynamics. We are constantly searching for
a suitable user interface, too. Stress will also be laid
on model of the quality verification and setting
model parameters using data from real social
networks. The mid-term goal is also to implement in
more detail the emotional states of agents.
ACKNOWLEDGEMENTS
This work was supported by internal grant of the
Institute of Technology and Business in České
Budějovice, grant No. 1/2013.
REFERENCES
Ahmad, M. A., Teredesai, A., 2006. Modeling spread of
ideas in online social networks. In: Proceedings of the
Fifth Australasian Conference on Data Mining and
Analystics - Volume 61 (Sydney, Australia, November
29 - 30, 2006). Australian Computer Society,
Darlinghurst, Australia.
Balint, M., Posea, V., Dimitriu, A., Iosup, A., 2011. An
analysis of social gaming networks in online and face
to face bridge communities. In Proceedings of the
third international workshop on Large-scale system
and application performance (LSAP '11). ACM, New
York, NY, USA.
Federico, P., Aigner, W., Miksch, S., Windhager, F.,
Zenk, L., 2011. A visual analytics approach to
dynamic social networks. In Proceedings of the 11th
International Conference on Knowledge Management
and Knowledge Technologies (i-KNOW '11), Stefanie
Lindstaedt and Michael Granitzer (Eds.). ACM, New
York, NY, USA.
Jelínek, J., 2011. Modeling of information flow in social
networks (in Czech) In: CEUR Workshop
Proceedings, Vol-802, http://ceur-ws.org/Vol-802/
(online), ISBN: 978-80-248-2369-0.
Ma, H., Huang, J., 2013. CUT: community update and
tracking in dynamic social networks. In Proceedings
of the 7th Workshop on Social Network Mining and
Analysis (SNAKDD '13). ACM, New York, NY,
USA.
Menges, F., Mishra, B., and Narzisi, G., 2008. Modeling
and simulation of e-mail social networks: a new
stochastic agent-based approach. In: Proceedings of
the 40th Conference on Winter Simulation (Miami,
Florida, December 07 - 10, 2008).
Palazuelos, C., Zorrilla, M., 2012. Analysis of social
metrics in dynamic networks: measuring the influence
with FRINGE. In Proceedings of the 2012 Joint
EDBT/ICDT Workshops (EDBT-ICDT '12), Divesh
Srivastava and Ismail Ari (Eds.). ACM, New York,
NY, USA.
Siebers, P.O. and Aickelin, U., 2007. Introduction to
Multi-Agent Simulation. In: Encyclopedia of Decision
Making and Decision Support Technologies. IDEAS
Group.
Wasserman, S. and Faust, K., 1994. Social Network
Analysis. Cambridge University Press, ISBN
0521382696.
Wu, W., Zhang, Y., 2010. Stability analysis in dynamic
social networks. In Proceedings of the 2010 Spring
Simulation Multiconference (SpringSim '10). Society
for Computer Simulation International, San Diego,
CA, USA.
Zhao, X., Sala, A., Wilson, Ch., Wang, X., Gaito S.,
Zheng, H., Zhao, B., 2012. Multi-scale dynamics in a
massive online social network. In Proceedings of the
2012 ACM conference on Internet measurement
conference. ACM, New York, NY, USA.
InformationDisseminationinSocialNetworks
271