Modeling of Information Sharing on the Business Social Media

Abstract

Social media, which support communication on the Internet, continue to spread rapidly. SNSs and micro-blogs are typical services. This research proposes methods for efficient information sharing through social media, especially intra-firm systems for information sharing of research or expertise. In this research, a simple model of social media is provided, whose users share information by directly referring to other users. The authors focused on a particular intra-firm social media and analyzed the usage of its communities. By applying a clustering method, they are classified into six groups and analyzed their characteristics and occupation rate. Finally, the authors proposed a method that supports information sharing and evaluated it by agent-based simulation using our proposed model. The authors clarified that not only encouraging users to post articles but also giving them an opportunity to refer to the existing community' s history is effective for information sharing.

Full text

SICE Journal of Control, Measurement, and System Integration, Vol.6, No. 2, pp. 083–087, March 2013
Modeling of Information Sharing on the Business Social Media
Fujio TORIUMI ∗, Takashi OKADA ∗∗, Shuichiro YAMAMOTO ∗∗∗, and Kenichiro ISHII ∗∗∗
Abstract : Social media, which support communication on the Internet, continue to spread rapidly. SNSs and micro-blogs
are typical services. This research proposes methods for efficient information sharing through social media, especially
intra-firm systems for information sharing of research or expertise. In this research, a simple model of social media is
provided, whose users share information by directly referring to other users. The authors focused on a particular intra-
firm social media and analyzed the usage of its communities. By applying a clustering method, they are classified into
six groups and analyzed their characteristics and occupation rate. Finally, the authors proposed a method that supports
information sharing and evaluated it by agent-based simulation using our proposed model. The authors clarified that not
only encouraging users to post articles but also giving them an opportunity to refer to the existing community’ s history
is effective for information sharing.
Key Words : social network services, information sharing, agent-based simulation, social media.
1. Introduction
As the steady growth of new network communication tools
continues, the expansion of social media such as social net-
works services (SNS), and blogs are becoming a social phe-
nomenon impacting societies all over the world. For example,
Facebook1, the largest SNS, has more than 500 million active
users [1]. Also, twitter2, micro-blog site, has more than 170
million users [2]. Social media have also received attention
from many companies, because they are expected to be a good
space for information sharing. Therefore, in this simulation, a
SNS model to analyze its influence on information sharing is
proposed.
In general, the purpose of social media is not always informa-
tion sharing. Thus, a model of a company SNS that shares in-
formation for product development is proposed. This research
proposed methods for efficient information sharing through so-
cial media, especially intra-firm systems used for information
sharing of research or expertise.
The authors analyze a social medium managed in a enterprise
to clarify how social media are used in real business settings to
propose a model of social media. Simulations are performed
with the social media model to clarify its effective features to
realize efficient information sharing.
2. Related Studies
Previous work analyzed the relation between knowledge
sharing in organizations and weak ties [3]. In this study, hu-
man relationships which are called weak ties [4] in sociology,
are more effective for knowledge sharing than such weak con-
∗Graduate School of Engineering, The University of Tokyo,
Japan
∗∗ NTT Information Sharing Platform Laboratories, Japan
∗∗∗ Graduate School of Information Science, Nagoya University,
Japan
E-mail: tori@sys.t.u-tokyo.ac.jp
(Received July 9, 2012)
(Revised September 27, 2012)
1http://www.facebook.com/
2http://twitter.com/
nections as those between departments.
Other work also analyzed the influence of the social net-
works of workers on knowledge sharing [5], concluding that
both ties and organization structures influence knowledge shar-
ing. Strong ties impact knowledge sharing.
On the other hand, there have been many previous studies of
online social media. Adamic et al.[6], for example, studied a
university SNS called Nexus and analyzed its structure and the
attributes and personalities of its users. Yuta et al.[7], who in-
vestigated the network structure of Mixi, discovered a gap in
its community-size distribution that is not observed in real so-
cial networks and developed a simple model to account for this
feature. Ahn et al.[8] compared the structures of three online
SNSs, (Cyworld, MySpace, and orcut), each with more than 10
million users. Twitter has also been analyzed [9],[10].
Many researches have addressed modeling and simulations
of knowledge sharing or emergent innovations. N. Gilbert et
al. modeled product development in the organizations [11] and
the structures of academic science [12]. Y. Fujita et al. studied
the roles of innovators in information distribution using simu-
lations [13]. Other studies modeled SNSs [14],[15] to simulate
human behaviors.
3. Proposal for Social Media Model
3.1 Overview
In this paper, the authors modeled a SNS called the “Chie-
no-Wa SNS” managed by the NTT Information Sharing Plat-
form Laboratories. The NTT group utilizes a community site
named “Chie-no-Wa,” which can be translated as knowledge
harmony, through its subsidiary companies. Chie-no-Wa SNS
is one subsystem of Chie-no-Wa. This SNS was developed to
share information. Communities and forums are its main fea-
tures. Therefore, the authors will model it as information shar-
ing sites using the community system using multi-agent based
modeling to model the social media.
3.2 Social Media Model
3.2.1 Agent
We define N agents that share the information in an SNS as
JCMSI 0002/13/0602–0083 c
2012 SICE

SICE JCMSI, Vol. 6, No. 2, March 201384
ai(i=1,2,···,N). Each agent has an information array that
is represented as Ndimensional vector Ei=(ei1,ei2,···,eiN ).
Each element eij takes binary value {0,1}.
All agents have specific objective functions calculated from
the information they possess. By repeating information shar-
ing, all agents maximize their objective function values. Com-
pared to product development, this process resembles informa-
tion collection to find and develop suitable products. Therefore,
agents affiliated with the same groups have the same objective
functions. The details of objective function will be explained in
Section 3.3.
Our model has two types of agents. “Information provider
agents” provide information to other agents. “Information
browser agents” rarely provide information. The number of in-
formation provide agents are decided by the analysis results of
a real social network service shown in Section 4.
3.2.2 Community
A community is a system to make groups in a SNS to find
other users who have the same interests. All users can write
information to the Bulletin board system in communities and
read the information written by other users of the Bulletin board
system. To represent this community feature, we modeled it as
an information repository . The agents in the model participated
in several communities. All agents can browse the accumulated
information provided by other agents from the communities in
which they participated. From the above procedure, all agents
maximize their objective function value. The following are the
agent behaviors for the community:
•Information provision
•Information browsing
The details of all behaviors are explained in Section 3.4. The
outline of information sharing by community is shown in Fig. 1.
Fig. 1 Information sharing model by community.
3.3 Objective Function using NK-Landscape Model
In our proposed model, all agents participating in the social
media have their own objectives. The quality of information to
achieve such objectives is represented by an objective function.
A higher objective function value represents a high quality of
information to develop feasible products. The objective func-
tion is modeled using a NK-Landscape model [16]. When we
assume information Eias one genotype, the fitness of the gene
Table 1 Sample of fitness array.
000 0.141
001 0.592
010 0.653
011 0.589
100 0.793
101 0.233
110 0.842
111 0.916
is assumed to match the objective function value. In the NK-
Landscape model, the fitness of the genes is evaluated by the
value of each locus and its neighbors. The fitness of gene Eiin
the NK-Landscape is evaluated by the following equation:
fi(Ei)=1
N
j
fij(eij,z(j)
i1,···,z(j)
iK ),(1)
where each locus eij takes 0 and 1, and z(j)
ik represents the
loci which have relationships with eij. The fitness of a lo-
cus is derived from its value and the related loci. Related loci
z(j)
i1,···,z(j)
iK are the Ks that surround the locus of loci eij.When
j=3,K=2, the following elements are used to calculate the
fitness of eij:
{eij,z(j)
i1,···,z(j)
iK }={ei2,ei3,ei4}.(2)
Considering that each locus eitakes a binary value, array
{eij,z(j)
i1,···,z(j)
iK }becomes a K+1lengthbitarray.
In the NK-Landscape model, we prepare fitness arrays for the
bit arrays. A sample of the fitness array for K=2isshownin
Table 1. When the bit array for locus eij is {eij,ei(j−1),ei(j+1) }=
{010}, the fitness of eiis calculated from the fitness array as
follows:
fij(eij,ei(j−1),ei(j+1) )=fij(010) =0.653.(3)
The fitness of the gene is calculated from the average of each
locus.
Each locus have its own fitness array generated by 2K+1ran-
dom value that takes [0.0,1.0]. Considering that each locus has
adifferent fitness array, to evaluate all of the gene loci, a fitness
table comprised of N2K+1random values is generated before-
hand.
3.4 Agent Behavior
3.4.1 Information provision
All agents provide their information to the community at a
certain rate. This is a representation of the written informa-
tion to the BBS in a community in real social media. The fol-
lowing are the information provision procedures. First, we de-
cide whether the provision has occurred in certain steps from
the provision rate calculated from the amount of provided in-
formation in the latest 30 steps. The detail of the information
provision is explained in Section 5.1. Next, if the provision is
determined to have occurred, the agent who provided the in-
formation is selected. Information provider agents are selected
first. When the information provider agents are not selected,
the other agents are selected randomly. Finally, information Ei,
which selected agent ai, is provided to the community. The in-
formation provided by the agent is retained for a certain period.
While the information is being retained, the other agents who
access the community can browse it.

SICE JCMSI, Vol. 6, No. 2, March 2013 85
Table 2 Classification results.
Feature Vectors C1C2C3C4C5C6
Pm32.41 17.25 4.18 0.03 0.03 0
Ub15.21 89.49 12.61 35.22 1.94 6.57
Up3.88 7.33 1.99 0.03 0.03 0
Pmax 18.49 6.58 2.78 0.03 0.03 0
Occupancy 0.028 0.01 0.072 0.18 0.288 0.423
3.4.2 Information browsing
The agent selects one piece of information from the browsed
information, which is the latest Prinformation that is provided
to the participating community. The tournament selection [17],
which selects the best information from a few random bits of
selected information, selects the information.
3.4.3 Information learning
The agent learns the selected information. The learning pro-
cedure is represented by the idea of crossover in the Genetic Al-
gorithm (GA). A piece of information is assumed to be a gene.
By crossovering3two pieces of information, the agent generates
new information. If the new information has higher fitness, the
agent updates his/her information as new information.
4. Community Analysis
4.1 Classification of Community
To realize a realistic simulation, the authors analyzed real
social media. By using analysis results, the authors realized
realistic parameter settings for the simulation to improve its re-
liability. Since many types of social media communities exist,
the authors classified them by their features and clarified how
they are used in social media to design a reliable communica-
tion model.
4.2 Feature Vector
We classified communities on a SNS called “Chie-no-Wa
SNS,” managed by the NTT Information Sharing Platform Lab-
oratories. To classify the communities, we employed the K-
Means classification algorithm [18] and analyzed all types of
communities classified by it from the viewpoint of communi-
cation through communities.
The following feature vectors are used for classification :
•Number of posted articles:Pm
•Number of participating users:Ub
•Number of information providers:Up
•Number of articles posted by the most active user:Pmax
The information providers are the users who post articles to
the community. The normalized value of each feature vector is
used to solve the algorithm.
4.3 Classification Results
The feature vectors of types of communities C1,···,C6,
which were obtained from the classification, are shown in Ta-
ble 2.
Users in the community belonging to type C1often post ar-
ticles. The number of articles is the largest, and almost 25.5%
of the users post articles. On the other hand, half of the articles
3Two-point crossover is adopted.
are posted by the most dedicated user: the active user in such
communities.
In C2type communities, many users participated. The rate at
which articles are posted by the most active user is smaller than
C1. Also, the number of users who posted articles exceeds C1,
showing that in the C2type communities, many users actively
communicated with each other. In C3type communities, there
is some communication but less than in the C1type communi-
ties. The rate of information providers is only 15.7% in such
communities. In C4type communities, the average number of
participants is larger than C1type communities. However, no
articles were posted to the community; in other words, the uti-
lization rate of such communities is extremely small. C5andC6
types of communities have fewer participants and fewer posted
articles. Thus, these communities are not being used.
In the next section, we will simulate the effect of information
sharing on social media using classification results.
5. Simulation of Information Sharing by Community
5.1 Simulation Aim and Setting
The aim of this simulation is to represent the information
sharing by communities on a social medium.
To realize a realistic simulation, we used the result of com-
munity analysis that was explained in Section 4 as parameter
settings. Every community represented in the simulation has
the following parameters:
•Number of posted articles Pm
•Number of participating users Ub
•Number of information providers Up
•Rate of articles posted by the most active user Pmax
The number of posted articles Pmrepresents the number of
articles posted to the communication in 30 steps. In all steps,
the ratio at which articles are posted to the community is Pm/30.
The number of participating users Ubrepresents how many
agents participated in the community. The number informa-
tion providers Uprepresents the number of users who posted
articles to the community. Such users are decided in the prepa-
ration phase, which never changes during the simulation. On
the other hand, all agents can browse the information from the
community.
The rate of articles of the most committed user Pmax repre-
sents how many articles are posted by the active user. When
Pmax is large, the community is led by the active user. On the
other hand, when Pma x is small, debates often occur in which
many users participated. Using the above parameters, we mod-
eled the six types of communities (Table 2). The types of com-
munities are randomly set to C1−C6based on the occupancy.
The other simulation settings are show in Table 3.
5.2 Evaluation
The simulation result was evaluated by the fitness averages of
the information calculated from the objective function. When
the fitness increases, the mechanism of information sharing by
community is effective for product development.
5.3 Base Simulation Results
First, we calculated the base simulation that has the above
parameters. The simulation result is shown in Fig. 2. The x-axis

SICE JCMSI, Vol. 6, No. 2, March 201386
Table 3 Simulation settings.
Agents M1000
Communities C150
Browsable information Pr10
Objective function La5
Length of information N20
Related locus K5
Tournament size St10
Simulations 100
Fig. 2 Base simulation results.
shows the simulation steps, and the y-axis shows the average
objective function values of all agents. The result is the average
of 100 simulations.
The simulation result clearly shows that for the first 10 steps,
the average value of the objective function increased rapidly.
However, after 20 steps passed, the increment rate of the ob-
jective function value decreased. Although the maximum ob-
jective function value is 1.0, the final objective function value
is insufficient. Therefore, in the next section, we seek a more
effective method to support information sharing.
6. Simulation of Efficient Method for Information
Sharing
6.1 Simulation Aim and Settings
The simulation aim is to find a method to support efficient
information sharing. The simulation parameters are set to the
same simulation explained in Section 5.
In this simulation, the following two methods are proposed
to support information sharing. The first method is large com-
munity, and the second is community recommendation.Theva-
lidity of all methods are checked by simulation.
The authors proposed these two methods for the following
reasons. The social media we modeled was developed to realize
information sharing in companies. In such cases, since methods
that require drastic system changes are problematic, we chose
methods that only require slight change for systems.
6.2 Information Sharing Support Methods
6.2.1 Large community
In this method, a large community in which all users partic-
ipate is created in the social media model. In this community,
an article is posted by following all the steps.
This model represents the introduction of a company-wide
community. Such a community often encourages inter-
department communications. Also, with large communities, the
amount of information that agents can browse is expected to in-
Fig. 3 Simulation results.
crease.
6.2.2 Community recommendations
In this method, a community is recommended to all agents
at a certain rate (10% here). When the agent recommends a
community, participation increases. This model represents the
community recommendations from other users. For example,
when a department requires new technologies to develop prod-
ucts, its members recommend their own communities to other
company members who have knowledge of the technologies to
share the information of new technologies.
The following two types of recommendations are simulated
here:
•To agents with the same objective function
•Randomly to agents
6.3 Simulation Result and Discussion
The simulation result is shown in Fig. 3. The x-axis shows
the simulation steps, and the y-axis shows the averages of the
objective function values of all agents. Additionally, the simu-
lation calculated in Section 5 is shown as a base line result.
From Fig. 3 all methods that support information sharing ef-
fectively improve the average fitness of the information of all
agents. In other words, both methods are effective for informa-
tion sharing support. Different characteristics are confirmed in
each change in the fitness line.
The large community result shows that fitness increased dras-
tically in the first few steps, but became saturated in the early
periods because all agents were given a chance to find various
users, which improved the information fitness. However, the
information that can be found in large communities does not
always elevate the information fitness of agents. On the other
hand, the community recommendations realized smooth but not
saturated improvement of information fitness. Finally, the fit-
ness of the community recommendations became larger than
that of the large communities.
From the difference between the results of recommendations
to an agent with the same objective function and random rec-
ommendations to an agent, objective function has a greater af-
fect than information sharing. However, random recommenda-
tions are also affective. Thus, recommendations seem to be an
extremely effective method to support information sharing. In
the recommendation methods, the fitness of the agent informa-
tion improved despite no increments of article posts because the
agents have more opportunities to find information that was al-
ready provided to the community. Such additional information

SICE JCMSI, Vol. 6, No. 2, March 2013 87
provides chances to elevate the fitness of the information.
From the simulation result, we predict that users must contact
much information to raise its value. Therefore, introducing a
new system that recommends other communities in which users
are not participating effectively supports information sharing.
For example, one implementation might be a system that finds
other users with knowledge about the topics when the commu-
nity discussions become deadlocked.
7. Conclusion
In this research, the authors provided a simple model of so-
cial media, whose users are sharing their information by a com-
munity. To realize realistic parameters for the model, the au-
thors focused on a particular intra-firm social medium and an-
alyzed the usage of its communities. By applying a K-Means
clustering method, we classified them into six groups and ana-
lyzed their characteristics and percentages.
Finally, the authors proposed methods that support informa-
tion sharing and evaluated them by an agent-based simulation
using our proposed model. The authors clarified that not only
encouraging users to post articles but also giving them the op-
portunity to refer to the existing community’ s history is effec-
tive for information sharing.
Future work includes modeling other social media and im-
plementing our proposed methods. To enhance the cogency of
each model including leaning process is one of the most impor-
tant future works.
Acknowledgments
This research was conducted under the collaboration scheme
between Nagoya University and NTT. The authors would like
to thank the scheme and people who supported it.
References
[1] Inc. Facebook: Statistics — facebook,
http://www.facebook.com/press/info.php?statistics
[2] Inc. Twitter: About twitter, http://twitter.com/about
[3] M.T. Hansen: The search-transfer problem: the role of weak
ties in sharing knowledge across organization subunits, Admin-
istrative science quarterly, Vol. 44, No. 1, pp. 82–85, 1999.
[4] M.S. Granovetter: The strength of weak ties, The American
journal of sociology, Vol. 78, No. 6, pp. 1360–1380, 1973.
[5] R. Reagans and B. McEvily: Network structure and knowledge
transfer: The effects of cohesion and range, Administrative Sci-
ence Quarterly, Vol. 48, No. 2, pp. 240–267, 2003.
[6] L.A. Adamic, O. Buyukkokten, and E. Adar: A social network
caught in the Web, First Monday, Vol. 8, No. 6, p. 29, 2003.
[7] K. Yuta, N. Ono, and Y. Fujiwara: A Gap in the Community-
Size Distribution of a Large-Scale Social Networking Site,
Arxiv preprint physics, 0701168, 2007.
[8] Y.Y. Ahn, S. Han, H. Kwak, S. Moon, and H. Jeong: Analysis
of topological characteristics of huge online social networking
services, Proceedings of the 16th international conference on
World Wide Web, pp. 835–844, 2007.
[9] A. Java, X. Song, T. Finin, and B. Tseng: Why we twitter: un-
derstanding microblogging usage and communities, Proceed-
ings of the 9th WebKDD and 1st SNA-KDD 2007 workshop on
Web mining and social network analysis, ACM New York, NY,
USA, pp. 56–65, 2007.
[10] H. Kwak, C. Lee, H. Park, and S. Moon: What is Twitter, a
social network or a news media?, Proceedings of the 19th in-
ternational conference on World wide web, ACM, pp. 591–600,
2010.
[11] N. Gilbert, A. Pyka, and P. Ahrweiler: Innovation networks: a
simulation approach, Journal of Artificial Societies and Social
Simulation, Vol. 4, No. 3, 2001.
http://ideas.repec.org/a/jas/jasssj/2000-25-1.html
[12] N. Gilbert: A simulation of the structure of academic science,
Sociological Research Online, Vol. 2, No. 2, 1997.
http://ideas.repec.org/a/sro/srosro/1997-15-1.html
[13] Y. Fujita, Y. Washida, F. Toriumi, K. Ueda, and K. Ishii: Re-
versal of influence: decrease of innovator’s influence under in-
formation diversification, Proceedings of the 9th International
Conference on Autonomous Agents and Multiagent Systems,
Vol. 1, pp. 1543–1544, 2010.
[14] F. Toriumi, K. Ishida, and K. Ishii: “Encouragement Methods
for Small Social Network Services,” IEEE/WIC/ACM Interna-
tional Conference on, Web Intelligence and Intelligent Agent
Technology, 2008, Vol. 1, pp. 84-90, 2008.
[15] T. Fujio and I. Kenichiro: Simulation of Encouragement Meth-
ods for SNS based on User Behavior Model, Proceedings of
3rd World Congress on Social Simulation, 2010.
[16] D.A. Levinthal: Adaptation on rugged landscapes, Manage-
ment science, Vol. 43, No. 7, pp. 934–950, 1997.
[17] B.L. Miller, B.L. Miller, D.E. Goldberg, and D.E. Goldberg:
Genetic algorithms, tournament selection, and the effects of
noise, Complex Systems, Vol. 9, pp. 193–212, 1995.
[18] J. MacQueen: Some methods for classification and analysis
of multivariate observations, Proceedings of the Fifth Berke-
ley Symposium on Mathematical Statistics and Probability,
pp. 281–297, 1967.
Fujio TORIUMI
He received Ph.D. from Tokyo Institute of Technology
in 2004. He joined Nagoya University as Assistant Pro-
fessor in 2004. And he moved the University of Tokyo as
Associate Professor. Recently, he is engaged in research
of social media analisys and agent-based simulation.
Takas h i O KADA
He received B. Eng. and M. Eng. degree from the Uni-
versity of Tokyo in 1987 and 1989 respectively. Since
1989, he is working for NTT (Nippon Telegraph and
Telephone Corp.) laboratories. Recently, he is engaged in
practical research of knowledge management and imple-
mentation of the knowledge management process based
on new generation communication system in the NTT
group.
Shuichiro YAMAMOTO
He joined NTT in 1979. He moved NTT Data Cor-
poration in 2002. And he moved Nagoya University in
2009. He is the leader of SEC Dependable System Tech-
nology WG of IPA. He received academic achievements
awards from IPSJ in 2002 and IEICE in 2003.
Kenichiro ISHII
He received M. Eng. degree from the University of
Tokyo in 1974 and he joined NTT. He received D.Eng.
degree from the University of Tokyo in 1990. And he
moved Nagoya University in 2003 as Professor.