J. Software Engineering & Applications, 2009, 2: 283-287
doi:10.4236/jsea.2009.24036 Published Online November 2009 (http://www.SciRP.org/journal/jsea)
283
Research on the Trust Model Based on the
Groups’ Internal Recommendation in
E-Commerce Environment
Nan REN1, Qin LI2
School of Economics and Management, Jiangsu University of Science & Technology, Jiangsu, China.
Received June 30th, 2009; revised August 2nd, 2009; accepted August 7th, 2009.
ABSTRACT
The trust plays an extremely important role in online shopping. In order to make online shopping trusty, this paper puts
foreword a new trust model in e-commerce environment GIR-TM (Groups’ Internal Recommendation Trust Model).
First, it regarded the network as a combination of groups, and then did the internal recommendation based on these
groups. The GIR-TM, in the process of recommendation, distinguished clearly between the trust degrees of recommen-
dation node and the trust degrees of recommended node, and then calculated the integrated credibility value of the
recommended node according to the weight of recommendation node in the group, the partial trust degree and the de-
gree of recommendation when the recommendation node recommends the recommended node, and the overall credibil-
ity value of recommended node as well. Lastly through listing the experimental data and comparing with the HHRB-TM
(History and Honest Recommendation Based Trust Model) on the same condition, it is verified that GIR-TM is feasible
and effective.
Keywords: E-commerce, Groups, Internal Recommendation, the Credibility Value
1. Introduction
Trust is the basis of co-operation, and it plays an ex-
tremely important role in online shopping. At present,
there are still some issues in peer-to-peer trust model,
such as over-reliance on the recommendations of others,
trust calculations’ inaccuracy, difficulty of dealing with
the united malicious attacks, dynamic strategy malicious
nodes and so on [1]. In order to promote e-commerce to
develop stably and quickly, researchers around the world
have done some researches in the field of trust model: M.
H. Hanif Durad et al. [2] discussed how to utilize the
trust management to strengthen its security in the grid
environment. G. Liang et al. [3] discussed how to use the
trust management system such as reputation systems to
solve the trust problem among the users. F. Almenarez et
al. [4] discussed how to use the auto-negotiation tech-
nologies to solve the dynamic trust management in gen-
eral environment. Jøsang A et al. [5] pointed out that in
the P2P environment, the core problem of the trust
mechanism based on the reputation were that: in the
given application, what trust factors are the most appro-
priately used to infer the measurement of trust and repu-
tatition? How to generate, acquire and aggregate the inform-
ation about these trust factors? Whether the trust mecha-
nisms can resist various attacks which are controlled by
strategic individuals? Li Wen [6] put forward a History
and Honest Recommendation Based Trust Model in
Peer-to-Peer Networks, and improved the evaluation al-
gorithm of trust. Chen Xiaoliang [7] classified the impact
factors, built the calculation model of initial trust, and
figured out that Consumers had distrust of web sites and
online stores which is a bottleneck in e-commerce de-
velopment.
To solve the core problem that consumers are lacking
in trust of e-commerce currently, this paper establishes a
trust model based on the groups’ internal recommenda-
tion in E-commerce environment, in which the compre-
hensive credibility value of recommended node is calcu-
lated by the weight of recommendation node in the group,
the partial trust degree when recommendation node rec-
ommends recommended node, and the degree of recom-
mendation and the overall credibility value of recom-
mended node. The calculated result can supply the basis
for restraining malicious acts effectively (such as joint
defamation, malicious exaggeration, providing false in-
formation, etc.).

Research on the Trust Model Based on the Groups’ Internal Recommendation in E-Commerce Environment 284
2. GIR-TM
2.1 Group Mechanisms
2.1.1 Group’s Structure
First of all, according to the credibility value of every
node, the peer-to-peer network can be divided into three
small collections [8]:


,, BadGoodWholeN 
In which: Good is a collection of nodes with good
credibility value gained through the good service pro-
vided to others.
Bad is a collection which is composed of malicious
nodes;

is a collection of nodes whose credibility are un-
sure.
When a node p joins in the network, it is put into the
collection

, and its credibility value is 0. The nodes
with good performance can increase their credibility
value into a particular value until it is placed into the
collection Good. On the contrary, if the node performs
badly and its credibility value will be less than 0, then it
will be moved to the collection Bad, meanwhile, the in-
formation about its bad performance will be notified in
the whole network. As received the notice, the nodes will
not trade with the notified node any more and then cut
off the connection with it.
Then, the nodes in collection Good will be grouped.
Some nodes in this collection have a certain credibility
value, higher reliability and stability, which compose the
group called Trusted Group (TG), and we assume that its
scale is Q. When the credibility value of a node in collec-
tion Good reaches a certain degree, it can set up its own
trust group or apply to join in the existing groups. Those
nodes that have not joined in the TG are put into another
group (AG).
The administrator of a TG is a node which creates the
group initially. Administrators must maintain a connec-
tion with all the nodes inside of group, and we assume
that each node in group maintains k as external connec-
tion. Administrators can choose which node to join in,
while the node can also choose its trust group. In order to
clarify the information of each trade and the credibility
value of each node in trading, we suggest that it is nec-
essary to establish a Node’s Information (NI) for each
node in the net to record its own series of activities, just
as shown in Table 1.
After a node establishes its own TG, the node needs to
notify its information to the nodes which do not belong
to any TG and the administrators of other trust groups
(TGs). After the administrator of another TG receives the
notice, it will inform the message to the nodes in its own
group.
According to the above strategy, the entire network is
divided into n TGs, collection Bad, collection

, as well
as AG. As shown in Figure 1, we assume that all nodes in
collection G ood have entered the trust group, just as the
two trust groups TG1 and TG2 in Figure 1, and A, B re-
spectively represent the administrators of TG1 and TG2,
and the number of their foreign connection is K (K = 1,
2 ... n).
2.1.2 Co n nection of N odes
After the node’s credibility value in collection

reaches
a certain trust degree (the credibility value of collec-
tion

R

) through the good behavior, it will be moved into
collection Good according to the principle "two-way
choice", that is, the node can choose trust group, and the
administrator selects a node, while the node can choose
to stay in AG, join in or build a TG.
As shown in Figure 1, through transacting with other
nodes, node C’s credibility value satisfies RC, then
it can enter the collection Good. When the node C de-
cides to join in the TG1, it needs to send application in-
formation to the administrator A, the information in-
cludes its ID and the kinds of commodities.After re-
ceiving the application, the administrator A must carry
out the following steps:

c
S

R
Step1: First of all, calculate the number of the nodes in
TG1, if the number reaches Q, reject the node C’s con-
nection, otherwise continue to Step 2;
Step2: Judge that whether the node c is a malicious
node or not, if it is, refuse its connection, otherwise con-
tinue to Step 3;
Step3: The administrator reviews the node C’s credi-
bility value, 1
TG
R
is the credibility value of TG1, if
1>, reject the node C’s connection, other-
wise continue to Step4;
TG
Rc
R

R
Step4: The node p belongs to TG1, if p TG1, the
kinds of commodities of p satisfies, the node C
is allowed to joining in, continue to Step5;

c
S
p
S
Table 1. The Node’s Information (NI)
Transaction node
Group
ID
Group
Administrator
Value of
credibility Node’s
ID
Number of
successful
Number of
failure
Credibility
Value
Integrated credibility
value
……
……

Research on the Trust Model Based on the Groups’ Internal Recommendation in E-Commerce Environment 285
Step5: The administrator allows the node C to join in
TG1, and establishes a connection with the node C;
Step6: C creates connection with all the nodes in this
group and its own NI table.
After C receives the refused news, it can choose other
TG, and repeat the above steps. If it is rejected by all
administrators, it can set up its own group or stay in AG.
2.1.3 Depart u re of No des
Nodes’ departure has two ways: one is active departure,
and the other is passive departure. Active departure can
withdraw from the peer-to-peer actively when the node
completes the transactions. If the node is also the admin-
istrator, before it leaves, it will choose the node with the
highest credibility value in the group as administrator,
and copy the information of the group to it. Passive de-
parture happens when a node’s credibility value is less
than the credibility value of the group, and the adminis-
trator ejects it out of the trust group, and puts it into col-
lection Bad.
2.2 Internal Recommendation Mechanisms
2.2.1 Recomm e nda ti o n Ide a s
As same as the real life, and on the basis of the Trust
Group, the basic framework of GIR-TM is established, as
shown in Figure 2,
In the Figure 2, the closed circle equals to a TG, in
which each node (A, B, and C…) has its own transaction
nodes. In order to clarify the information of each transac-
tion and the credibility value of each transaction node,
we proposed to establish a table called Node's Informa-
tion (table NI) for every node. The table NI includes the
ID, administrators of the group where the node stays, and
the credibility value of node, as well as the transaction
information such as the ID of other transactions nodes,
the number of successful transactions and one of failed
transactions, the credibility value of other nodes and the
integrated credibility value (the calculation of the inte-
grated credibility value is referred to the following sec-
tion) after transacting. Assuming that A has transacted
with the node E and with the completion of each
Figure 1. Netw or k schematic of trust group-based
Figure 2. Basic framework of groups’ internal recommen-
dation trust model
transaction, the node A will refresh its own NI table (this
NI table is to be shared, the shared region is the inte-
grated credibility value with a recommendation tip sign),
and then through comparing to judge whether the inte-
grated credibility value of E meets the requirements or
not, that is, if the integrated credibility value of E is
greater than the overall credibility value of the group
which includes node A, it will be signed with the rec-
ommended tip, which will be shared by the other nodes
in this group, otherwise, giving up their recommendation.
2.2.2 Calculation of the Integrated Credibility Value
In order to calculate the Integrated Credibility Value of
GIR-TM in Figure 2, we firstly introduced the following
3 definitions [9] about the partial trust, the degrees of
recommendation, and the integrated credibility value:
Definition2.1 partial trust: represents the par-
tial view of the node U on node V, which directly comes
from the historical transaction experience between them,
given that
vu
L
Nuv
S
LUV
vu   (1)
In which, represents the number of successful
transaction with node V in view of U;
uv
S
uv
N represents the number of total transactions be-
tween U and V within the last interval timet

(t

is
illustrated that the trust model pay more attention to the
time limit of the nodes’ behavior). If = 0, then
= 0.
uv
N
vu
L
Definition 2.2 the degrees of Recommendation: the
degrees of recommendation how node X recommends the
node V is calculated as follows:
xvxv
xvxv
vx FS
FS
R


 (2)
In which, Fuv represents that in node U’s view, the
number of failure transaction with node V. If Sxv + Fxv =0,
then set=0. Or if Sxv - Fxv <0, then set=0. As
shown in definition 2.2, if the nodes perform badly in the
trading, and gain poor assessment from others, then his
vx
Rvx R

Research on the Trust Model Based on the Groups’ Internal Recommendation in E-Commerce Environment 286
credibility value will not be increased, but drastically
reduced. So to some extent, this way can restrain the ma-
licious acts of malicious nodes.
Definition 2.3 the integrated credibility degree: set
representing the projection of the trust level
which node U trusts node V (U
vu
G

V) in the trust scope λ:
vu
G= [

vu
L+ (1-

) ]
v
T

(3)
Here Tv is the overall credibility degree of the node V,
Tv <1. In which, α is a constant and 0 <α <1, and λ is the
trust scope, λ > 1.
Based on the quantitative description of the trust men-
tioned above, Document [6] put forward a HHRB-TM
(History and Honest Recommendation Based Trust
Model) in Peer-to-Peer Networks, and the corresponding
integrated trust credibility value is calculated as the For-
mula (4):
vu
G=u

vu
L+ (1- u

) () (4)  



N
i
vx
R
1
x
Cr
In which, is the credibility value of node X,
x
Cr u

[10] is a weight factor about node U referring to its own
direct history transaction experience. u

is dynamic, and
changes with the time or the number of transactions.
But the Formula (4) does not consider the weight of
recommendation node, and the role of the node with high
credibility value does not play completely. Aiming at
such problem, this paper put forward the definition of
comprehensive weights, and the calculation formula is:
Definition 2.4 comprehensive weights: set as
comprehensive weights of a trust group, and the calcula-
tion formula is:
x
Gt
x
Gt =


m
i
i
x
Cr
Cr
1
(5)
Here is the credibility value of node X,
is the sum of the credibility value of all the nodes
in the group in which node X is included.
x
Cr


m
i
i
Cr
1
According to the principle of the higher credibility
value, the higher credibility, the more accurate recom-
mendation information, and then the bigger contribution
rate, the weight of recommendation node is added into
the calculation of the credibility value of recommended
node, just as:
v
Ct = (6)
x
Gt vx Rx Cr
In which, is comprehensive weight of the trust
group in which the recommendation node X is included;
x
Gt
vx
R is the recommending evaluation that recom-
mendation Node X puts foreword on the recommended
node V;
x
Cr is the overall credibility value of recommended
node.
v
Ct is gained by the feedback information about the
Node V’s recommendation, 1;
v
Ct 
Combining the partial trust with recommendation trust
through (7):
vu
G’= u

vu
L+(1- u

) (7)
v
Ct
In which, ’is the integrated credibility value to
represent how the node U recommends node V;
vu
G
vu
L represents accumulated direct history transac-
tion experience when the node U transacts join with node
V;
So, the integrated credibility value is as follows:
vu
G’= u

vu
L+ (1-u

)x
Gt vx R

x
Cr (8)
3. Model Validations
We compared HHRB trust model with the trust model
GIR in this article in the same experiment situation. In
order to verify the model, enumerating 20 groups of ex-
perimental data, and assuming =380 ，=430，
uv
Suv N
u

=0.9.
These 20 groups of experimental data include the
number of successful transactions, the number of
failure transactions and the credibility value of each
node . Each group of experimental data is listed ran-
domly, because the GIR-TM is based on the TG. We
could think the nodes in the group are reliable, and the
failure rata is smaller, and then we could abide the prin-
ciple that the listed data of the failure transactions num-
ber is always less than the successful transactions num-
ber, in addition, the failure transactions number needs to
be much less. And the of each node needs to be
more than 0.5 (because every node is in the TG and
should have a higher credibility value, and we assume
that the threshold value of each TG’s credibility value is
0.5). As the real life, the reputable people will form a
group, and they are all reliable to have much more possi-
bility to transact each other successfully. In generally,
these experimental data are realistic and are shown in
Table 2:
xv
S
xv
F
x
Cr
x
Cr
The experimental results in Figure 3 are calculated
according to Table 2. As shown in Figure 3 (GN is group
number), although the integrated credibility value of the
HHRB model fluctuates sometimes, the integrated credi-
bility values obtained by the HHRB model and the GIR
model are more or less the same, just between 0.8825
and 0.8838. So it can be concluded that the GIR-TM is
verified to be feasible and effective.

Research on the Trust Model Based on the Groups’ Internal Recommendation in E-Commerce Environment 287
Table 2. Experimental data
S
xv F xv Crx S xv F xv Crx S xv F xv Crx S xv Fxv Crx
1 250 3 0.6 6 79 1 0.91 11 31 0 0.65 16 283 2 0.95
2 60 1 0.7 7 301 3 0.68 12 1730 0.92 17 291 3 0.98
3 80 1 0.57 8 123 2 0.73 13 92 1 0.835 18 68 1 0.74
4 99 0 0.73 9 161 1 0.88 14 1822 0.72 19 136 0 0.9
5 59 1 0.82 10 83 1 0.79 15 2592 0.935 20 197 1 0.54
Figure 3. Comparing the integrated credibility value be-
tween and
vu
Gvu G
4. Summary and Prospect
The paper puts forward a Trust Model Based on the
Groups’ Internal Recommendation by analyzing the cur-
rent issue of trust in electronic commerce. The model is
composed of Trust Group and internal recommendation
mechanism. Generally speaking, the achievements are as
follows:
1) Put forward the GIR model based on the TG;
2) Put forward the algorithm of the integrated credibil-
ity value of the GIR model by improving the algorithm
of the integrated credibility value of the HHRB;
3) Verify the effectiveness of the GIR model by com-
paring it with HHRB model.
Theoretically, the model can provide a good trading
environment for customers, and reduce the occurrence of
malicious actions. However, besides effectiveness veri-
fication of the GIR model, the model needs to be verified
in the following aspects:
1) Verify the fairness of the transaction and the accu-
racy of the algorithm described in the model;
2) Further improve the model according to tested re-
sults;
3) Based on this paper, study on the rewarding and
punishment mechanism to reward the reputable people
and punish the malicious node.
5 Acknowledgments
This research was supported by Qing Lan Project of Ji-
angsu Province of China.
REFERENCES
[1] X. Li and L. Liu, “A reputation-based trust model for
peer-to-peer e-commerce communities[C],” Proc of IEEE
Conference on E-Commerce, ACM Press, California, pp.
275–284, 2003.
[2] M. H. Hanif Durad and C. Yuazlda, “A vision for the
trust managed grid [A], Proceedings of the sixth IEEE
International Symposium on Cluster Computing and the
Grid [C], IEEE Computer Society Washington, DC, USA,
2006.
[3] G. Liang, L. Junzhou, and X. Yaobin, “Developing and
managing trust in peer-to-peer systems [A],” proceedings
of the 17th International Conference on Database and
Expert Systems Applications[C], IEEE Computer Society
Washington，DC，USA，2006.
[4] F. Almenarez, A. Marin, D. Diaz，and J. Sanchez, “De-
veloping a model for trust management in pervasive de-
vices [J],” 2006.
[5] A. Jøsang, R. Ismail, and C. Boyd, “A survey of trust and
reputation systems for online service provision [J],” De-
cision Support Systems, Vol. 43, No. 2, pp. 618−644,
2007.
[6] W. Li, “A history and honest recommendation based trust
model in peer-to-peer Networks [D],” Hunan University,
pp. 28–29, September 2007.
[7] X. L. Chen, “Research on the modeling and application of
initial trust for e-commerce[D],” Huazhong University of
Science and Technology, 2008.
[8] A. Gummadi and J. P. Yoon. “Modeling group trust for
peer-to-peer access control,” Database and Expert Sys-
tems Applications, Proceedings 15th International Work-
shop, pp. 971–978, 2004.
[9] W. Dou, H. M. Wang, and Y. Jia, “A recommenda-
tion-based peer-to-peer trust model [J],” Journal of soft-
ware, Vol. 15, No. 4, pp. 571–583, 2004.
[10] G. Zacharia and P. Maes, “Trust management through
reputation mechanisms [J],” Applied Artificial Intelli-
gence, Vol. 14, No. 9, 2000.