Information Sciences 278 (2014) 250266

Contents lists available at ScienceDirect

Information Sciences
journal homepage: www.elsevier.com/locate/ins

Evolution of knowledge sharing behavior in social commerce:

An agent-based computational approach
Guoyin Jiang a,b,, Feicheng Ma b, Jennifer Shang c, Patrick Y.K. Chau d
a
School of Information Management, Hubei University of Economics, Wuhan 430205, China
b
School of Information Management, Wuhan University, Wuhan 430073, China
c
Katz Graduate School of Business, University of Pittsburgh, Pittsburgh, PA 15260, United States
d
School of business administration, The University of Hong Kong, Hong Kong, China

a r t i c l e i n f o a b s t r a c t

Article history: The rapid development of e-commerce has expedited knowledge growth in the e-com-
Received 8 July 2013 merce social community. Knowledge sharing among online users has exhibited a nonlinear
Received in revised form 30 October 2013 dynamic evolution. This paper examines the evolutionary process of knowledge sharing
Accepted 13 March 2014
among users of the social commerce; builds an evolutionary game model to depict knowl-
Available online 21 March 2014
edge sharing phenomenon in the virtual community; and develops a mixed learning algo-
rithm based on individual users historical game strategy, neighborhood users strategy,
Keywords:
and information noise. We design a computational model based on multi-agent theory
Social commerce
Knowledge sharing
and social network, and implement computational experimental system using NetLogo
Dynamic evolution 5.0. We nd that the proposed computationalexperimental model can help decision mak-
Computational experiment ers simulate evolutionary process under various scenarios. The evolutionary game rule and
social network structure signicantly inuence the degree of cooperation and knowledge
sharing among users. The greater noise the network information has the less stable the
users behavior will be. One can thus identify an optimal initial cooperation rate to
facilitate the system to reach equilibrium state quickly. Our study on the dynamic evolu-
tion of knowledge sharing behavior in the social commerce contributes to the theoretical
development of literature and provides valuable decision-making support to managers.
2014 Elsevier Inc. All rights reserved.

1. Introduction

The emergence of Web 2.0 has changed the way users and enterprises interact and collaborate with each other. It creates
social commerce by combining social media, social network and e-commerce. Social commerce is lled with social interac-
tions and users in the online commerce community and other social network rate and share online product information and
advice [1]. Individual opinions earnestly contributed to the online community have enhanced knowledge sharing and
steered new consumer behavior.
iResearch (www.iresearch.com) reports that the online shopping behavior of social media users have signicantly
changed in recent time, as (i) the community of seeking friends online is growing rapidly; (ii) social media users pay more
attention to acquaintance, and are more condent of friends comments as they are often skeptical of advertising claims;
and (iii) the integrated e-commerce and social media can re-sort the users social relationships, and effectively motivate the

Corresponding author at: School of Information Management, Hubei University of Economics, Wuhan 430205, China. Tel.: +86 27 81972191.
E-mail address: guoyin_jiang@126.com (G. Jiang).

http://dx.doi.org/10.1016/j.ins.2014.03.051
0020-0255/ 2014 Elsevier Inc. All rights reserved.
 G. Jiang et al. / Information Sciences 278 (2014) 250266 251

product spread and form virtuous circle [2]. In particular, iResearch revealed that 54% of social media users respond that they
would purchase a product if their friends like it, as opposed to 28% of the overall e-shoppers.
Social commerce exemplies a collaborative environment where people share information and resources, so that the total
effect is greater than the sum of individual effects. The users in social commerce can inuence the buying behavior of others
by word of mouth, recommendation and transfer of knowledge on the social network [3]. Examples of social commerce can
be found in. Pinterest (www.pinterest.com), ebay community (community.ebay.com), and taobao froum (bbs.taobao.com).
Users can quickly access and readily capture other peoples experiences and derive comprehensive and relevant information
about the merchandise. In short, knowledge exchange has inundated the social commerce and effective knowledge manage-
ment is crucial for enhancing e-commerce performance. However, there is a dearth of research directed toward optimizing
knowledge sharing in social commerce [4]. Few studies have empirically examined what drives continued knowledge shar-
ing [5,6], but none have explored the evolution issues from the perspective of mathematical modeling and simulation ap-
proach in social commerce.
Registered users in social commerce build their networks by publishing shopping and answering shopping posts. Users
can easily see the remarks, reply, and comments of other users. For example, a registered user in the discussion forum can
quickly locate the posts of connected users after log-on. Users may share real knowledge, reserve or even distort facts, and
even post incomplete or inaccurate information. A user often becomes more active over time; their remarks and replies in-
crease network knowledge and before long the network would form a nonlinear evolutionary state. Users can choose knowl-
edge sharing strategies (actual experience or unreal shopping information) based on their own preference and others
actions. They may choose a strategy now, adopt a different one later, and use a combination strategy at the end. The e-com-
merce administrator thus should decide on what management strategies suit the network structure and user behaviors best.
The knowledge sharing behavior among users in a social commerce exemplies the evolution of knowledge interactions
over time. In this research, we employ agent-based computational approach to explore the dynamic evolutionary process of
knowledge sharing among users of the social commerce. The goals of this study is to (1) model user interactions in the social
commerce using game theory and identify the equilibrium of the dynamic evolution; (2) build a computational model to
study the behavioral evolution of knowledge sharing in social commerce based on evolutionary game, multi-agent, and so-
cial network theory; and (3) analyze how management strategies affect e-commerce discussion forums, and how to select
knowledge sharing network structures; examine how network parameters and information transparency shape the evolu-
tionary equilibrium. In short, we seek to provide mechanisms for social commerce administrators to establish knowledge
sharing strategy and manage under different scenario.
Our paper is organized as follows. In Section 2, we review the extant literature. Section 3 proposes a theoretical model and
develops a dynamic methodology to help reach evolutionary equilibrium. In Section 4, we conduct computational study to
validate the proposed model and understand the evolution of knowledge sharing in e-commerce. Section 5 presents compu-
tational results through virtual experiment, while the implications, limitations, and suggestions for future work are given in
Section 6. Section 7 concludes this research.

2. Literature reviews

2.1. Virtual community and social commerce

In social commerce, people communicate with each other through electronic media [7]. The virtual community comprises
people, intention of sharing, policy, and computer system [8]. Bhattacherjee [9] denes a virtual community as one gathering
rational community users, interacting with virtual spaces, and sharing among users. Therefore, obtaining knowledge through
information collection and user interactions is a key feature of a virtual community. The e-commerce forum offers a special
virtual community, which can integrate social features and enhance e-commerce performance.

2.2. Knowledge sharing in virtual community

Literature on knowledge sharing in a virtual community abounds. Most researchers have examined the drivers affecting
the participation in the community, and found individual motives include internal, external, and social motives [1012].
These factors affect the initial motivation of users joining the community for knowledge sharing [11,12]. Yet, the initial mo-
tives may not continue. Few studies focus on the inuence of trust instead of motives on users continuous participation of
the virtual community [5]. In fact, social learning takes place during the virtual community involvement, and users may alter
their motivation. Sun et al. [6] nd that task complexity and self-efcacy (two social learning factors) moderate the relation-
ship between motivation and sustained participation. The transactional complexity and the exterior motive have a negative
correlation, whereas complexity and the interior motive have a positive non-linear relationship.
Existing studies assume users continuing participation in virtual community is monocyclic and static [5,6]. However, the
continuous knowledge accumulation, expansion and interaction is a dynamic evolutionary process. Its evolutionary strategy
has a signicant inuence on organizational performance. An optimal strategy to knowledge evolution signicantly aid the
accumulation and improvement of knowledge [13]. Few literature on users continuous participation is based on the theory
252 G. Jiang et al. / Information Sciences 278 (2014) 250266

of social relationship [14]. As knowledge sharing creates value, investigating knowledge sharing in a virtual community from
the economic perspective is a tting attempt to explain the drive of knowledge sharing in social commerce.

2.3. Modeling knowledge sharing through game theory

Samieh and Wahba [15] studies knowledge sharing in the virtual community through game theory and shows that the
payoff for knowledge sharing among individuals can be described by multi-party game. Shih et al. [16] claims that learning
type and stimulus affect knowledge sharing in a cooperative team. Bandyopadhyay and Pathak [17] build an interactive
game model between the contractee and the contractor, and conclude that sharing complementary knowledge could boost
performance. Cai and Kock [18] use game to model e-collaborative evolution and identify strategies that would improve joint
performance. Ding and Huang [19] concludes, through a game framework, that collaboration is benecial to knowledge cre-
ation, while certain injection ratio of leaders and followers may render unstable collaboration. Li and Jhang-Li [20] nd that
knowledge benets can be improved by IT investment and stimulus mechanism, while decision-makers preferences and
information integrity affect knowledge sharing.
Information shared in social commerce may be true, overstated, or misleading. If users can build a trusting and cooper-
ative relationship, the information published or recommended may be trustworthy. If they deviate from the tenet of knowl-
edge sharing, the information may be deceptive and hurt social commerce users. Prior research has shown that network
topology and social factors signicantly impact consumers decision-making process [21]. Choi et al. [22] investigated
how network structure impacts the dynamics of innovation diffusion, while others show that different network structures
(e.g. small-world, random, and scale-free) have distinct inuences on cooperative behavior [2325]. Nowak and May [26]
study evolutionary game structure based on prisoners dilemma and conclude that a regular lattice facilitates the cooperative
behavior [26]. Hauert and Doebeli [27] show that the spatial topological structure in the snow drift game inhibits the emer-
gence of cooperative behaviors.
In these studies, the users are dened as nodes and every node in the network has only one state. It holds one state to-
ward its neighboring node either via cooperation or defection. However, in social commerce, each agent (user) interacts with
multiple users and implements different strategies toward them under different circumstances. So an agent may choose a
cooperation strategy toward neighbor A, and pick a defection strategy against neighbor B [26,28,29].

2.4. Investigating the evolutionary process based on computational method

Computational experiments, such as multi-agent model, help one to understand the complexity, dynamics and adaptive-
ness of a knowledge sharing system [30]. They are well-suited to study evolutionary behavior [31,32], as they describe the
autonomy, interaction, reactivity and reasoning behavior of complex systems. Traditional game theories generate macro-le-
vel (general) conclusions, while the evolutionary game and agent-based computational experiments we adapted in this
study would bring micro level (specic) results to the social commerce study.
The traditional evolutionary game method models macro-level evolution of games between two populations: one holds a
cooperation strategy and the other adopts a defection strategy. However, the evolutionary game model cannot represent
individual interactions. A well-suited approach for modeling micro-level individual interactions is the agent-based simula-
tion. Through detailed modeling, the agent-based simulation can ultimately provide informative macro-level summary and
offer valuable managerial implications.
In this research, we employ a unied computational experiment framework by integrating multiple agents, evolutionary
game, and social network to study the cooperation and defection strategies among users under different game rules and net-
work structures. By examining the path strategy (the combined past strategies) and evolutionary equilibrium, we contribute
to the behavioral evolution theory of knowledge sharing in the virtual community and offer decision support for manage-
ment practice in social commerce.

3. Theoretical model for knowledge sharing in social commerce

3.1. The game theory based model

Knowledge sharing strategies among users are affected by the experience they accumulated, their perception of other
users knowledge sharing behaviors, and changes in the external setting. The behavior of knowledge sharing in the social

Table 1
Game payoff matrix with penalty parameters.

Player 1 Player 2
Cooperation Defection
Cooperation b  c/2, bc/2 b  c, b  d
Defection b  d, b  c, 0, 0
 G. Jiang et al. / Information Sciences 278 (2014) 250266 253

commerce can be expressed by the interactive game relationship. Table 1 gives a symmetric game payoff matrix, which in-
cludes a social punishment.
In Table 1, the b stands for the benet of knowledge sharing, c is the expense of knowledge sharing, while d is the loss of
giving trust to awed information in knowledge sharing. For instance, in ebay community a recognized fault can be pub-
lished, as evidenced in the community content policy: http://pages.ebay.com/help/policies/user-created-content-ov.html.
The transformation of knowledge sharing behavior in social commerce indeed is a game process evolved in the user group
over time based on the game payoff matrix.

3.2. Identifying evolutionary equilibrium through dynamic replication

The evolutionary game equilibrium can be detected by the dynamic replication method [33,34]. This method can provide
theoretical support for management to make macroscopic decisions on knowledge dispersion, regardless of network struc-
ture and environmental factors.
Let P be the percentage of community users holding a cooperative attitude. Then, the expected cooperation benet from
knowledge sharing can be expressed as:
Ec Pb  c=2 1  Pb  c 1
The expected benet of defection from knowledge sharing is Ed:
Ed Pb  d 2
The average benet of this group is E:

E P  Ec 1  P  Ed 3
From dynamic replication, we nd the state of the evolutionary system over time can be expressed as:
dP h c  i
PEc  E P1  PEc  Ed P1  P P d  b  c  b 4
dt 2
Therefore, the evolutionary system has three possible solutions to game equilibrium.
 
cb cb
P 0; 1; c ; and P c 5
2
db 2
db
From Eqs. (3)(5), we are able to determine the equilibrium of the game under different scenarios. We summarize them in
Propositions 1-3 below.

Proposition 1. If any of the following scenario conditions are met, the group evolutionary equilibrium will reach defection.
(c (c
2
db>0 2
db>0
Scenario 1 ; i:e:
P P 1 b < c; c P 2d
(c (c
2
db<0 2
db<0
Scenario 2 ; i:e:
P60 b 6 c; c > 2d
8c 8c
>
> db >0 >
> db>0
<2 <2
Scenario 3 0 < P  < 1 ; i:e: b < c < 2d
>
> >
>
: :
P0 < P P0 < P

Proof. See Appendix A.1. h

Proposition 2. If any of the following scenario conditions are satised, the group evolutionary equilibrium will reach
cooperation.
(c (c
2
db<0 2
db<0
Scenario 4 
; i:e:
P P1 b > c; c 6 2d
(c (c
2
db>0 2
db>0
Scenario 5 
; i:e:
P 60 b P c; c < 2d
254 G. Jiang et al. / Information Sciences 278 (2014) 250266

8c 8c
<2 d  b > 0
> <2 d  b > 0
>
Scenario 6 0 < P < 1 ; i:e: b < c < 2d
>
: >
:
P0 > P P0 > P

Proof. See Appendix A.2. h

Proposition 3. If the conditions in Scenario 7 are satised, we will nd the equilibrium is at P = P.

c c
db<0 db<0
Scenario 7 2 ; i:e: 2
0 < P < 1 b > c > 2d

Proof. See Appendix A.3. h

Using the dynamic replication method (see Appendix A) to solve evolutionary game for knowledge sharing in social com-
merce can help design the knowledge sharing mechanisms in a macro-environment, regardless of the network structure and
information noise. However, in a knowledge sharing game, the evolutionary result is not a simple accumulation of all game
states held in a single time period. The historical game path, interactive users decisions and information noise, need to be
taken into account comprehensively. Such evolution has non-linear features. Thus, the evolutionary equilibrium cannot be
simply determined by the traditional dynamic replication method. It has to be analyzed by the computationalexperimental
method from the perspective of agency theory, game theory, and social network.

4. The computational model

4.1. Combination of game strategies among social commerce users

When an agent in the social network interacts with a neighbor user during a time period of knowledge sharing, she may
hold a cooperation or defection strategy. She may follow a different strategy in the next time period. The choice of strategy
depends on her own as well as the decisions of other users. Fig. 1 shows the eight possible strategy paths. FS(t) is the strategy
chosen by the user at term t. NS(t) is the strategy chosen by her neighbor user at time t. Depending on her own strategy and
her neighbors strategy choice, she will pick one of the eight strategy paths for time t + 1.

4.2. The game learning algorithm for users in the social commerce

The strategy chosen by an agent at time t depends on the strategy of the neighboring users at time (t  1) and the agents
own strategy. Learning algorithms for the evolutionary game include random process and heterogeneous group mixed
learning algorithm [35]. The learning algorithm chosen in this study resembles the learning method seen in the dynamic rep-
lication method. That is, a game player will imitate the behavior of the other game player, who attained the highest prot in
the previous time period. Note that we do not examine the imitation of single strategy a user employed but to consider the
path of strategies she adopted thus far.

FS(t)

Cooperation Defection

NS(t) NS(t)

Defection Defection
Cooperation Cooperation

FS(t+1) FS(t+1) FS(t+1) FS(t+1)

Defection Defection Defection Defection

Cooperation Cooperation Cooperation Cooperation

CCC CCD CDC CDD DCC DCD DDC DDD

Fig. 1. The eight path strategies from time t to t + 1.

 G. Jiang et al. / Information Sciences 278 (2014) 250266 255

In the knowledge sharing game, an individual learns from the neighbor with the highest benet, following the stochastic
process expressed in Eq. (6):
1
Pj ! i 6
1 euj ui =k
where uj is the benet generated by agent j and it is the maximum among those accrued by neighborhood users, while ui is
the accumulated benets of agent i and k is the information noise [36]. When k is very large (very noisy), say approaching 1,
P will approximate , equivalent to the result of coin-tossing. On the other hand, when k is very small (e.g. approaching 0),
the values of P will be close to 1, indicating agent i is sure to imitate user j.

4.3. Percentage of cooperation and defection

In our study, a node in the network is either in a cooperation- or defection-state. We dene the % of cooperation in a
network as

sum of the nodes holding the cooperation strategy

% of cooperation  100%;
2  total number of connections
and the % of defection in a network as

sum of the nodes holding the defection strategy

% of defection  100%
2  total number of connections

4.4. Formalize the social commerce

4.4.1. Dene the dimensions of a social commerce network

Newman [37] maintains that a complex network such as a social network possesses the small-world features, where most
users are not direct neighbors of one another, but they can be reached from a small number of hops or steps. By applying data
mining techniques to the data collected from Taobao (the largest Chinese e-commerce rm) forum, we nd that the social
commerce does share the features of a small-world and the nature of a scale-free network, which has a degree distribution
following a power law [37,38].
The two sets of data we collected from the Taobao forum include one with 554 nodes (Sample A), and the other with 1073
nodes (Sample B). Using Ucinet 6, we attain the following statistical characteristics:

(1) Average path length. An average path length is the average number of steps between node i and node j, along the shortest
paths dij. It is a measure of the efciency of information or mass transport on a network [39].

1 X
L1 dij 7
2
NN 1 iPj

where N is the number of nodes in the network. A small L indicates that information can be communicated more directly and
easily.
Most real life networks have a very short path length leading to the concept of a small world where everyone is connected
to everyone else through a very short path. For Samples A and B, the average path length is 3.026 and 2.356 respectively. We
found the path lengths are comparable and both samples are relatively small, and only 23 links will connect the agent to the
target user.

(2) Network clustering coefcient. In social commerce, nodes often intertwine and generate groups with high degree of ties. A
network clustering coefcient is a measure of degree to which nodes in a network tend to cluster together [39]. It is usually
between 0 and 1, and a fully connected network has a clustering coefcient of 1. We found the weighted clustering coef-
cients of our two samples are 0.829 and 0.790 respectively. Both are large and comparable.

(3) Degree distribution. The degree of a node in a network is the number of edges the node has connecting to other nodes. We
dene the degree distribution as the probability distribution of these edges over the whole network.
We found the maximum degree of Sample A is 43, while that of Sample B is 95. The minimum degree in each sample is 1.
Fig. 2 shows the degree distribution of the two samples, while Table 2 identies the most appropriate model to t the ob-
serve degree distribution. We found that the Power Model ts the degree distribution data best and no signicant differences
are found between the two samples.
Our analysis above shows that the social commerce has the feature of the small-world, which indicates it has a short
average path length and a large Network clustering coefcient; and its degree distribution obeys the power law. Due to these
three characteristics, we conclude that the network structure of the studied social commerce is a scale-free network.
256 G. Jiang et al. / Information Sciences 278 (2014) 250266

Fig. 2. Degree distribution of the two samples.

Table 2
Results of curving tting.

Equation Model summary Parameter estimates

R2 F df1 df2 Sig. Constant b1
Power .618 30.734 1 19 .000 36.069 1.156
Sample A
Exponential .292 7.850 1 19 .011 6.836 .069
Logistic .292 7.850 1 19 .011 .146 1.071

Power .558 42.986 1 34 .000 37.021 .962

Sample B
Exponential .184 7.670 1 34 .009 4.450 .024
Logistic .184 7.670 1 34 .009 .225 1.024

In view of the limited sample data we collected, we feel it may be necessary to examine the evolution of knowledge
sharing under the small-world network and random network in our research, in addition to the scale-free network structure
associated with our two samples.

4.4.2. Formalize the model denition

Let each node in the virtual network be an agent and there are n agents. We dene the community network N as N = {X,
ST, NT, NB, FI, F, t}, in which,

(1) X is the set of agents and X = {agent1, agent2, . . . , agentn}. Each agent is a user in the network.
(2) ST is the state space with ST = {ST1, ST2, . . . , STj, . . . , ST8}. Each STj, is a path strategy.
(3) NT is the set of network types and NT = {Smallword, Random, Scalefree}. These three network structures exemplify the
classic complex networks discussed in literature [37].
NT
(4) NB is the state space of all neighbors of an agent. NB = {NB1, NB2, . . . , NBn}, where NBi fagenti ! agentj g. Namely, NB
comprises users neighboring agent i in the NT network structure.
(5) FI is the set of benets derived by all agents, FI = {FI1, FI2, . . . , FIn}.
(6) F is the state transfer function. F: {(FIi, NBi) ? ST(t)}  t ? ST(t+1). The state of agent i at time t + 1 is a function of param-
eters, including its own prot, game strategy, and the strategies of other users who have direct connection with agent i
at time t.

5. Computationalexperimental system and virtual experiment

In this section, we design the computational system, simulate the virtual community and evolutionary process, collect
numerical results under different scenarios, and make comparisons through statistical analysis. The simulation and exper-
iments are implemented by NetLogo 5.0.
 G. Jiang et al. / Information Sciences 278 (2014) 250266 257

5.1. Experimental system and the default parameters

We build an agent-based model to explore the interactions among all individuals over time (200 time periods in our case).
The experimental system is divided into four modules:

(1) Set up the network, and dene the game parameters and the model parameters.
(2) Graphically illustrate the network evolution, and show how the nodes change dynamically in the network.
(3) Display the results, and show the characteristics of the network (e.g. clustering coefcient and path length).
(4) Draw the strategy evolution curve.

The parameters given in Table 3 are an example setting of scenario #1. Those values are determined by the conditions
shown under scenario #1 in Proposition 1. We can manually change the parameter settings when different simulation is
experimented to address a new scenario.

5.2. Experimental results under different game parameters and network structures

We study eight path strategies: CDC, DDD, CDD, DCD, CCD, CCC, DCC, DDC, as discussed in Fig. 1. Each agent in the network
connects with one or more agents (users). Each agent has his own game strategy toward different agents. For example, he
may have a cooperation relationship with the jth user in the neighborhood but have a defection strategy with neighbor k.
Therefore, each agent may employ one or more path strategies. Recall in Section 3.2 we divide the social commerce into se-
ven scenarios. We replicate each scenario 50 times (sample size of 50) so as to ensure the reliability of the experimental re-
sults. The length of each simulation cycle is set to 200 time periods. We explore the evolution of knowledge sharing using the
simulated data and compare the results with the theoretical results in Propositions 1-3. We rst employ ANOVA to test the
differences in means and variances at a 95% condence level. When p-value <0.05, we conclude there is signicant difference
between the tested populations. As we cannot guarantee that the simulated results are normally distributed, we also employ
the KruskalWallis H (a non-parametric) approach to test if samples are originated from the same population. When p-value
>0.05, we conclude that there is no signicant difference in population distributions.
For Scenario 1, to satisfy the condition: b < c, c > 2d, we assume b = 6, c = 10, and d = 4. The results are summarized in
Fig. 3. Fig 3(a), (c), and (e) shows the changes in the % of cooperation and defection over time, while Fig. 3(b), (d) and (f) display
the number of nodes holding the cooperation strategy under each path strategy in each network structures. From Fig. 3(a),
(c), and (e), we nd defection is the dominant strategy in the game, which can quickly reach equilibrium at time period 10.
The result of the simulation is consistent with the macro-trend found in Proposition 1. From Fig. 3(b), (d) and (f), we nd
path strategies CDC and CDD dominate others. The path strategy CDC is an amicable strategy combination, as the agent holds
a cooperation strategy at rst. Even though his opponent reacts with defection strategy, he still insists on applying the coop-
eration strategy subsequently. Path strategy CDD has a Tit for Tat relationship, where the agent will use cooperation strategy
at rst, then replicate opponents action and employ defection strategy. Path strategy CDD will affect users who opt to take
the defection initiative.
Under Scenario 1, the means of % cooperation on small-world, random, and scale-free networks respectively are 34, 34.37,
and 41.79. The p-value from ANOVA mean test is 0.00, indicating the means among the three network structures are signif-
icantly different. Alternatively, the standard deviations are 6.17, 5.66, and 7.08 respectively. The homogeneity test results in
a p-value of 0.00 (<0.05), indicating the variances are different among the three network structures.
The mean and variance of % of cooperation in the scale-free network are the largest. Through the KruskalWallis H test, we
nd the p-value is 0.00 (<0.05) and conclude that the results from the three network structures under Scenario 1 are
different.
Similarly, for Scenarios 27 we conduct simulation and perform statistical analyses. The results are summarized in
Table 4, which shows the mean and variance of % of cooperation (column #5), % of path strategies adopted (column #6),
and p-values for means, variances, and KruskalWallis H distribution tests (column #7). We nd that the dominant path
strategies (column #6, top three, boldfaced) are different under different scenarios and network structures.

Table 3
Default settings of the parameters in the experimental system.

No. Parameters Description Default

1 Num-nodes Number of nodes in the network 80
2 rewiring probability Reconnecting probability in small-world network 0.2
3 connection probability Connecting probability in a random network 0.02
4 theoretical unit Whether to apply P in eq. (5) to the network True
5 initial strategy Percentage of initial cooperation 0.5
6 b value of b in Table 1 6
7 c value of c in Table 1 10
8 d value of d in Table 1 4
258 G. Jiang et al. / Information Sciences 278 (2014) 250266

(a) % of cooperation & defection in small world network (b) path strategies in small world network

(c) % of cooperation and defection in random network (d) path strategies in random network

(e) % of cooperation and defection in scale-free network (f) path strategies in scale-free network

Fig. 3. % of cooperation and number of path strategies under three network structures in Scenario 1 (b < c, c > 2d).

Except for Scenarios 3 and 6, the variances under the scale-free structure are the largest, indicating that the game strategy
the agent employs is very unstable (volatile) in the scale-free network, as it has a highly asymmetric connection among
nodes. The larger variances a game strategy has, the more likely the group will have the Tit-for-Tat (retaliation) issue in
the knowledge sharing game, as evidenced in Scenario 4. When cooperation strategy is dominant, the mean % of cooperation
under scale-free network is the smallest, as seen in Scenarios 3, 4, 5 and 7. On the contrary, when defection is dominant, the
mean % of cooperation under scale-free network is the largest, as shown in Scenario 1.
 G. Jiang et al. / Information Sciences 278 (2014) 250266 259

Table 4
Statistical results of simulation data.

Scenario Expression Parameters Network Dominate strategy(mean, % Path strategy (CDC, DDD, CDD, p-Value
type variance of % of cooperation) DCD, CCD, CCC, DCC, DDC) (p1, p2, p3)
1 b < c, c > 2d b = 6, Small Defection(34, 6.17) 27.3, 0, 34.1, 0, 22.7, 15.9, 0, 0 0.00, 0.00,
c = 10, world 0.00
d=4 random Defection(34.37, 5.66) 23.1, 0, 37.5, 0, 20.1, 19.3, 0, 0
Scale-free Defection(41.79, 7.08) 25.9, 0, 45.1, 0, 12.8, 16.2, 0, 0
2 b = c, b  c/2  d > 0 b = 6, c = 6, Small Defection(36.731, 4.88) 0, 10.4, 0, 45.6, 0, 0, 20.8, 23.2 0.0, 0.06,
d=2 world 0.00
random Cooperation(52.8, 4.14) 0, 23.8, 0, 39.5, 0, 0, 20.5, 16.2
Scale-free Same(49.2, 5.77) 0, 25.7, 0, 33, 0, 0, 32.7, 8.6
3&6 b < c < 2d b = 4, c = 6, Small Cooperation(87.5, 7.06) 2.8, 2.3, 5.3, 1.2, 5.2, 47, 32.7, 3.5 0.00, 0.00,
d=5 world 0.00
random Cooperation(89.55, 7.4) 1.8, 2.8, 1.9, 2, 12.8, 56, 18, 4.7
Scale-free Cooperation(59.6, 2.19) 10.9, 11.4, 8.5, 11.4, 17.8, 12.2, 16.6,
11.2
4 (b  c) = (b  c/2  d), b = 6, c = 4, Small- Cooperation(87.79, 3.36) 7.1, 0, 1.1, 0, 38.1, 53.7, 0, 0 0.00, 0.00,
b  c/2  d > 0 d=2 world 0.00
random Cooperation(79.55, 2.94) 9, 0, 4.2, 0, 43.1, 43.7, 0, 0
Scale-free Cooperation(72.44, 13.38) 45, 0, 2.6, 0, 12.8, 39.6, 0, 0
5 b  c/2  d > 0 b > c, b = 7, c = 6, Small Cooperation(98.7, 8.52) 0, 0.3, 0, 0.2, 0, 0, 98.4, 1.1 0.00, 0.00,
c < 2d d=5 world 0.00
random Cooperation(87.99, 7.38) 0, 1.8, 0, 1.5, 0, 0, 77.6, 19.1
Scale-free Cooperation(82.43, 8.57) 0, 0.5, 0, 4, 0, 0, 66.7, 28.8
7 b > c > 2d b = 10, Small Cooperation(81.67, 2.98) 6.8, 0.2, 0.3, 5.4, 24.4, 32.1, 18.4, 0.00, 0.00,
c = 6, d = 2 world 12.4 0.00
random Cooperation(75.8, 4.38) 12, 0.4, 3.2, 8.4, 10.3, 33.8, 13.7,
18.2
Scale-free Cooperation(74.1, 6.98) 6.7, 0.2, 0.3, 5.4, 24.4, 32.1, 18.5,
12.4

Note: p1 is value from ANOVA, p2 is value from variance homogeneity test, and p3 is value from non-parametric test.

When b equals c, the probability of choosing either a defection or cooperation strategy is 0.5. But the punishment param-
eter d in game payoff can impact the evolution of game strategy. Despite b = c, in Table 4 we nd that the dominant strategies
and path strategies under Scenario 2 are very different in each of the three networks. Defection is a dominant strategy in the
small world network, while cooperation is a dominant strategy in the random network. But no dominant strategy exists in
the scale-free network, in which the mean % of cooperation is about the same as the % defection (the probability is approx-
imately 0.5 each). This suggests that in addition to parameters b, c and d, network structures may also impact the evolution
of game strategy. Moreover, in Section 3.2 we use the dynamic replication method to prove that the initial % of cooperation
(P0) impacts game equilibriums in Scenarios 3 and 6. Through simulation, we conduct tests to empirically examine the effect
of P0 on the equilibrium of game strategy.
Following the conditions of Scenarios 3 and 6 (i.e. b < c < 2d), we assume b = 4, c = 6, d = 5, which results in P = 0.5. When
P0 = 0.3 (P0 < P, corresponding to Scenario 3), we nd the dominant strategy is cooperation under each of the three network
structures. This result contradicts Proposition 1, as which is a macro level of analysis and does not consider the specic type
of network structure. On the other hand, when P0 = 0.8 (P0 > P, corresponding to Scenario 6), we nd the dominant strategy
is cooperation under each of the three network structures. This is consistent with Proposition 1.

5.3. Comparison of the experimental results under different game setting scenarios

Table 4 shows that different scenarios (parameter settings) have different % of cooperation under different network struc-
tures. Through ANOVA and the nonparametric test we obtain Table 5, in which the ANOVA tests reveal that the averages % of
cooperation among different scenarios are signicantly different. Similarly, the homogeneity tests show that the variances

Table 5
Test results under different scenarios in different network structure (at a 95% condence level).

Network structure p-Value of ANOVA p-Value of variance homogeneity test p-Value of nonparametric tests
Small-world network 0.00 0.00 0.00
Random network 0.00 0.00 0.00
Scale-free network 0.00 0.00 0.03
260 G. Jiang et al. / Information Sciences 278 (2014) 250266

Table 6
Test results under different initial cooperation percentage (at a 95% condence level).

Network Initial Mean and variance of % of p-Value of p-Value of variance p-Value of

structure percentage cooperation ANOVA homogeneity test nonparametric tests
Small-world 0.3 94.57, 5.9 0.00 0.01 0.00
network 0.5 87.5, 7.05
0.8 96.06, 8.08
Random network 0.3 73.69, 12.3 0.00 0.016 0.00
0.5 89.55, 7.4
0.8 74.27, 5.49
Scale-free 0.3 58.49, 2.15 0.00 0.00 0.03
network 0.5 59.6, 2.18
0.8 60.74, 3.59

Table 7
Test results at different levels of information noise (at a 95% condence level).

Network structure K Mean and variance of % of p-Value of p-Value of variance p-Value of nonparametric
cooperation ANOVA homogeneity test tests
Small-world 0 81.67, 2.98 0.00 0.056 0.00
network 1000 75.80, 4.38
10,000 74.10, 6.98
Random network 0 79.72, 3.76 0.00 0.00 0.00
1000 74.87, 3.94
10,000 72.76, 7.84
Scale-free 0 82.24, 4.90 0.04 0.00 0.009
network 1000 79.58, 3.97
10,000 72.31, 7.02

under the six scenarios are different in each network structure. We check the distribution differences of % of cooperation
between groups through the nonparametric test (the KruskalWallis H test), which uses ranks to determine if the samples
are from different distributions. We nd that they differ in different scenarios. In summary, different parameter settings,
corresponding to different scenarios in the social commerce, have signicant inuences on the % of cooperation.

5.4. Experimental results under different initial % of cooperation setting

We now test the impact of initial % of cooperation on overall % of cooperation. The initial percentages are set at 0.3, 0.5, and
0.8 respectively. The results are summarized in Table 6, from which we can conclude that the initial percentages of cooper-
ation have signicant inuences on % of cooperation.
From the study, we also nd different initial cooperation percentages converge at different speed. An optimal initial setup
percentage could steer the network to reach equilibrium rapidly.

5.5. Experimental result with incomplete information

We now examine the impact of information noise on the % of cooperation. The information noise levels are set at k=0,
1000, and 10,000 respectively. We summarize the results in Table 7, which concludes that different information noises have
signicant inuences on the % of cooperation connections, except for the homogeneity test of variances under the small-
world network. In fact, the mean value for k = 0, k = 1000, and k = 10,000 under the Small-world network are 81.67,
79.72, 84.24 respectively, and its variances are 2.98, 3.76, and 4.9, respectively. The greater the information noise is, the
higher the variance of % of cooperation, and the larger the uctuation in its % of cooperation over time.

6. Results and discussion

In this research, we study the evolutionary equilibrium of knowledge sharing among social commerce users through the
dynamic replication method. We analyze the behaviors of cooperation and defection in knowledge sharing among users in
the virtual community and the evolutionary process of strategy choice by using the agent-based computational approach.
We nd different game setups (scenarios) has a signicant inuence on users choice of path strategy and the cooperation
evolution of network. An efcient design of game payoff (penalty mechanisms d) can effectively regulate user behavior and
increase the % of cooperation in group game. We found that different scenarios have different % of cooperation in evolutionary
 G. Jiang et al. / Information Sciences 278 (2014) 250266 261

equilibrium and different converging speeds in reaching the equilibrium. A favorable mechanism can effectively regulate and
promote the knowledge sharing behavior among users in the virtual community.
Network structures have a signicant effect on the evolution of group cooperation and choice of path strategy. The % of
cooperation and the evolution of strategy combination vary with the network structures. Although cooperation or defection
may be a dominant strategy in a given scenario, the % of cooperation often varies signicantly under different network types.
Therefore, different reward and penalty measures should be adopted to guide the cooperation under different network struc-
tures, instead of taking up the same measure for all groups, i.e. communities of different network structure should adopt
different strategies.
From the experimental data, we nd that the mean % of cooperation under scale-free network is the smallest when the
cooperation strategy is dominant. On the contrary, the mean % of cooperation under scale-free network is the largest when
defection is dominant. Therefore, except for setting optimal game parameters, managers should also pay attention to the im-
pact of network structures on cooperation behavior. Possible effective strategies include applying positive guidance or neg-
ative impedance on key nodes or paths.
The initial cooperation ratio (P0) has a signicant inuence on the evolution of % of cooperation and choice of the path
strategy. We also nd different P0 will reach equilibrium at different speed. A good initial percentage would steer the group
game to arrive at equilibrium rapidly.
The information noise has a signicant inuence on the cooperation of knowledge sharing. The greater the information
noise, the higher the variance of % cooperation, and the larger the uctuation of evolutionary curve over time.
This study employs agent-based simulation method to explore the evolution of knowledge sharing behavior in social
commerce. It contributes to the literature in twofold: (1) Many studies have examined the motive of knowledge sharing
in virtual communities, but few have focused on the motive of continuous use and behavioral evolution. We develop a dy-
namic replication model for knowledge sharing among users in the virtual community from the perspective of evolutionary
game, and the model can be used to analyze the macro-level of group game. (2) We design a computational model that inte-
grates agent, evolutionary game, and social network. The model can help visualize the evolution of % of cooperation, defec-
tion and path strategy; grasp the dynamic evolution features in knowledge sharing game over time; and explore micro-level
interactions among users in different networks under various scenarios.
Four important managerial implications can be drawn for social commerce management:

(1) Different virtual communities (with different network structures) should adopt different macro-level managerial
strategies, i.e. customize the reward and penalty mechanisms. Different incentives and penalties can inuence users
choice of strategies and affect the performance of knowledge sharing.
(2) Managers should pay close attention to the initial motive of users, i.e. whether users are willing to cooperatively share
knowledge or not. An agents initial decision will affect the knowledge sharing behaviors of others and the entire com-
munity. Assessing and understanding the motivations of users are crucial.
(3) Intervention strategies should be applied to the virtual community when necessary. The forum moderator or manage-
ment should explain the policy regularly and guide the learning of knowledge with facts. For example, management
can provide objective data, graphics, direct quotes or video; place authoritative and highly reliable posts on top; grad-
ually carry out the evaluation and survey on the credibility of historical reports and the trustworthiness of user posts;
and properly reveal any relevant information. These methods will promote knowledge learning and give constraints to
key paths and nodes in the network. Typical examples include cutting off the communication of a few users, closing
some information of other users, and limiting the activities of users with poor credibility records.
(4) Providing awless channels for interactions will facilitate knowledge sharing and remove communication obstacles
among users. Also, one can add popular auxiliary devices to enhance the communication. For example, integrating
social media tools, such as Twitter, Alitalk, QQ, and blogs into the virtual community, will allow users to conveniently
and quickly interact with each other in the forum.

This paper has some limitations. First, we assume the network structure is xed, and we did not consider the evolution of
the network structure. However, the virtual community may dynamically change over time: some nodes may become dor-
mant, while other nodes may enter or even exit the network. Second, our study did not consider the personality differences
in individual users, e.g. trust, intimacy, and alienations. Third, the effects of different guides or impedances on key nodes and
paths are not analyzed. We shall add these elements into our future research, and offer theoretical foundation for dynamic
behavior evolution of knowledge sharing, as well as decision-making support for managerial practice.

7. Conclusions

The virtual community is an growing phenomenon of e-commerce. Designing effective managerial measures in the e-
commerce social community can facilitate knowledge sharing among users, promote e-commerce activities, and improve
e-business performance. This study examines the evolution of knowledge sharing behavior in social commerce based on
the dynamic replication method and agent-based computational approach, and draw conclusions that help managerial prac-
tice. In comparison with other methods, the proposed methods offer valuable macro- and micro-level insights on evolution
of knowledge sharing in social commerce. We nd that the rules (e.g. those stipulated by ebay community) of knowledge
262 G. Jiang et al. / Information Sciences 278 (2014) 250266

sharing game and social network structure signicantly affect the degree of cooperation and knowledge sharing among
users. The greater the noise information is in the network, the less stable the users behavior will be. An optimal ratio of ini-
tial cooperation can facilitate cooperation and help the system reach equilibrium state quickly.
The dynamic replication method can analyze macro-level evolutionary trend and approximate the equilibrium for group
game. On the other hand, the multi-agent simulation system for group game can visualize the interactions among users at
micro-level, and present macro-level insights about group behavior in the end. The ndings derived from the macro- and the
micro-level can help managers attain evolutionary data in different game settings and network structures. These data can aid
mangers in designing the best knowledge sharing mechanism and regulating the dynamic game environment. Agent-based
models have proven to be useful for understanding business dynamics. The advanced agent based computational approach
integrates both the agent-based model and the multidisciplinary models. The advanced agent-based computational ap-
proach integrates both the agent-based model and the multidisciplinary models. By combining the methodologies in eco-
nomics, social studies, and management, the proposed method bridges the gap between multiple elds, and offers a
valuable tool for researchers to conduct interdisciplinary studies.

Acknowledgements

We thank the editor-in-chief, the associate editor, two anonymous reviewers for their valuable comments and sugges-
tions on this work. This work was partially supported by a grant from the National Natural Science Foundation of China
(Nos. 71101047 and 70833005), China Postdoctoral Science Foundation (Nos. 2011M500119 and 2012T50674), and S&T pro-
ject of Hubei Provincial Department of Education (No. D20132201).

Appendix A

A.1. Proof of Proposition 1

We take dPdt
as the y axis and P as the x axis. In view of the distribution features of stable equilibrium points in Eq. (4),
Figs. A1 and A2 correspond to the conditions when P > 1 and P = 1, respectively. In Fig. A1, there are two stable equilibrium

dP/dt

0 1 P
P*

Fig. A1. P > 1 (scenario 1).

dP/dt

0 1 P
P*

Fig. A2. P = 1 (scenario 1).

 G. Jiang et al. / Information Sciences 278 (2014) 250266 263

dP/dt

0 1 P
P*

Fig. A3. P = 0 (scenario 2).

dP/dt

0 1 P
*
P

Fig. A4. P < 0 (scenario 2).

dP/dt

0 P* 1 P

Fig. A5. 0 < P < 1 and P0 < P (scenario 3).

dP/dt

0 P
1 P*

Fig. A6. P > 1 (scenario 4).

264 G. Jiang et al. / Information Sciences 278 (2014) 250266

dP/dt

0 P*
P
1

Fig. A7. P = 1 (scenario 4).

dP/dt

0 P
P* 1

Fig. A8. P = 0 (scenario 5).

dP/dt

P* 0 1
P

Fig. A9. P < 0 (scenario 5).

dP/dt

0 P* 1 P

Fig. A10. 0 < P < 1 and P0 > P (scenario 6).

 G. Jiang et al. / Information Sciences 278 (2014) 250266 265

dP/dt

0 P* 1 P

Fig. A11. 0 < P < 1 (scenario 7).

points at P = 0 and P = P. However, the initial values of the system are always within [1]. As P (>1) cannot be chosen as a
stable equilibrium point, only P = 0 is chosen as the stable equilibrium point. That is, all are defections. As to Fig. 2, appar-
ently only P = 0 appears at the stable equilibrium point.
Under Scenario 2, the stable equilibrium points are distributed as in Figs. A3 and A4. Similarly, the stable equilibrium
point appears at P = 0.
Under Scenario 3, the corresponding stable equilibrium points are as shown in Fig. A5. Two stable equilibrium points may
appear at P = 0 and P = 1. The choice of the values depends on the initial value of the system. If the initial value P0 < P, then
the stable equilibrium point is at P = 0. When P0 P P ; it is at P = 1. Therefore, under Scenario 3, the group evolutionary equi-
librium will reach defection.

A.2. Proof of Proposition 2

Under Scenario 4, Figs. A6 and A7 correspond to conditions when P > 1 and P = 1, respectively. Apparently, the stable
equilibrium point only appears at P = 1.
Under Scenario 5, the stable equilibrium point corresponding to P = 0, P < 0 is shown in Figs. A8 and A9. In Fig. A8, the
stable equilibrium point only appears at P = 1. In Fig. A9, there are two equilibrium points shown respectively at P = P, P = 1.
However the initial value of the system only falls within [0, 1]. Thus, the stable equilibrium point can only appear at P = 1.
Under Scenario 6, the corresponding stable equilibrium points are shown in Fig. A10. Two stable equilibrium points ap-
pear at P = 0 and P = 1 respectively. The equilibrium value depends on the initial value of the system. If the initial value
P0 < P, then the stable equilibrium point is at P = 0. When P 0 P P  , it is at P = 1. Therefore, under Scenario 6 the group evo-
lutionary equilibrium will reach cooperation.

A.3. Proof of Proposition 3

Fig. A11 illustrates the results under Scenario 7. Apparently, the only stable equilibrium point under Scenario 7 is at P = P.

References

[1] S.J. Yu, The dynamic competitive recommendation algorithm in social network services, Inf. Sci. 187 (2012) 114.
[2] C. Shi, Social Media have Higher Impact on Online shopping of Chinese Citizens, 2012. <http://ec.iresearch.cn/17/20120724/177375.shtml> (accessed:
10.03.13).
[3] Y.-M. Li, H.-W. Hsiao, Y.-L. Lee, Recommending social network applications via social ltering mechanisms, Inf. Sci. 239 (2013) 1830.
[4] J.-F. Wang, E-commerce communities as knowledge bases for rms, Electron. Commer. Res. Appl. 9 (2010) 335345.
[5] W. He, K.-K. Wei, What drives continued knowledge sharing? An investigation of knowledge-contribution and -seeking beliefs, Decis. Support Syst. 46
(2009) 826838.
[6] Y. Sun, Y. Fang, K.H. Lim, Understanding sustained participation in transactional virtual communities, Decis. Support Syst. 53 (2012) 1222.
[7] C. Romm, N. Pliskin, R. Clarke, Virtual communities and society: toward an integrative three phase model, Int. J. Inf. Manage. 17 (1997) 261270.
[8] J. Preece, Online Communities: Designing Usability and Supporting Sociability, Wiley, New York, 2000.
[9] A. Bhattacherjee, Understanding information systems continuance: an expectationconrmation model, MIS Quart. 25 (2001) 351370.
[10] R.M. Ryan, E.L. Deci, Intrinsic and extrinsic motivations: classic denitions and new directions, Contemp. Educ. Psychol. 25 (2000) 5467.
[11] C.-M. Chiu, M.-H. Hsu, E.T.G. Wang, Understanding knowledge sharing in virtual communities: AN integration of social capital and social cognitive
theories, Decis. Support Syst. 42 (2006) 18721888.
[12] M.-H. Hsu, T.L. Ju, C.-H. Yen, C.-M. Chang, Knowledge sharing behavior in virtual communities: the relationship between trust, self-efcacy, and
outcome expectations, Int. J. Hum. Comput. Stud. 65 (2007) 153169.
[13] D.-N. Chen, T.-P. Liang, Knowledge evolution strategies and organizational performance: a strategic t analysis, Electron. Commerce Res. Appl. 10
(2011) 7584.
[14] U. Lechner, J. Hummel, Business models and system architectures of virtual communities: from a sociological phenomenon to peer-to-peer
architectures, Int. J. Electron. Commerce 6 (2002) 4153.
266 G. Jiang et al. / Information Sciences 278 (2014) 250266

[15] H.M. Samieh, K. Wahba, Knowledge sharing behavior from game theory and socio-psychology perspectives, in: Proceedings of the 40th Annual Hawaii
International Conference on System Sciences, IEEE Computer Society, 2007, p. 187c.
[16] M.-H. Shih, H.-T. Tsai, C.-C. Wu, C.-H. Lu, A holistic knowledge sharing framework in high-tech rms: game and co-opetition perspectives, Int. J.
Technol. Manage. 36 (2006) 354367.
[17] S. Bandyopadhyay, P. Pathak, Knowledge sharing and cooperation in outsourcing projects a game theoretic analysis, Decis. Support Syst. 43 (2007)
349358.
[18] G. Cai, N. Kock, An evolutionary game theoretic perspective on e-collaboration: the collaboration effort and media relativeness, Eur. J. Oper. Res. 194
(2009) 821833.
[19] X.-H. Ding, R.-H. Huang, Effects of knowledge spillover on inter-organizational resource sharing decision in collaborative knowledge creation, Eur. J.
Oper. Res. 201 (2010) 949959.
[20] Y.-M. Li, J.-H. Jhang-Li, Knowledge sharing in communities of practice: a game theoretic analysis, Eur. J. Oper. Res. 207 (2010) 10521064.
[21] S.A. Delre, W. Jager, M.A. Janssen, Diffusion dynamics in small-world networks with heterogeneous consumers, Comput. Math. Org. Theory 13 (2007)
185202.
[22] H. Choi, S.-H. Kim, J. Lee, Role of network structure and network effects in diffusion of innovations, Ind. Mark. Manage. 39 (2010) 170177.
[23] B. Xianyu, Social preference, incomplete information, and the evolution of ultimatum game in the small world networks: an agent-based approach, J.
Artif. Soc. Soc. Simul. 13 (2010) 7.
[24] B. Xianyu, Other-regarding preference and the evolutionary prisoners dilemma on complex networks, Physica A 389 (2010) 11051114.
[25] B. Xianyu, Prisoners dilemma game on complex networks with agents adaptive expectations, J. Artif. Soc. Soc. Simul. 15 (2012) 3.
[26] M.A. Nowak, R.M. May, Evolutionary games and spatial chaos, Nature 359 (1992) 826829.
[27] C. Hauert, M. Doebeli, Spatial structure often inhibits the evolution of cooperation in the snowdrift game, Nature 428 (2004) 643646.
[28] K. Pakdaman, D. Mestivier, External noise synchronizes forced oscillators, Phys. Rev. E 64 (2001) 030901.
[29] O. Durn, R. Mulet, Evolutionary prisoners dilemma in random graphs, Physica D 208 (2005) 257265.
[30] L. Zhen, Z. Jiang, H. Song, Distributed recommender for peer-to-peer knowledge sharing, Inf. Sci. 180 (2010) 35463561.
[31] G. Jiang, B. Hu, Y. Wang, Agent-based simulation of competitive and collaborative mechanisms for mobile service chains, Inf. Sci. 180 (2010) 225240.
[32] G. Jiang, B. Hu, Y. Wang, Agent-based simulation approach to understanding the interaction between employee behavior and dynamic tasks, SIMUL.:
Trans. Soc. Model. Simul. Int. 87 (2011) 407422.
[33] J. Hofbauer, K. Sigmund, Evolutionary game dynamics, Bull. Am. Math. Soc. 40 (2003) 479519.
[34] J.M. Smith, G.R. Price, The logic of animal conict, Nature 246 (1973) 1518.
[35] B.S.K. AmirM, Another approach to mutation and learning, Games Econ. Behav. 14 (1996) 1943.
[36] W.B. Du, X.B. Cao, M.B. Hu, H.X. Yang, H. Zhou, Effects of expectation and noise on evolutionary games, Physica A (2009) 22152220.
[37] M.E.J. Newman, The structure and function of complex networks, SIAM Rev. 45 (2003) 167256.
[38] A.-L. Barabsi, R. Albert, Emergence of scaling in random networks, Science 286 (1999) 509512.
[39] D.J. Watts, S.H. Strogatz, Collective dynamics of Small-world networks, Nature 393 (1998) 440442.