A Fuzzy Trust Measurement Method for Mobile E‐Commerce

A Fuzzy Trust Measurement Method for Mobile E‐Commerce Zhengying Cai1,a*, Jun Chen1,b, Yijun Luo1,c College of Computer and Information technology, China Three Gorges University, Yichang 443002, China

1

*a

master_cai@163.com, b917136891@qq.com, c1932261722@qq.com

Abstract Today, lacking of trust and security is becoming a big obstacle to the application and postulation of the mobile commerce. Here the trust formation and operation of mobile commerce system are studied based on a fuzzy trust management model. Firstly, a trust evaluation model based on fuzzy mathematics is proposed, and related evaluated index system of mobile commerce system is built. Secondly, the fuzzy recommend and combined algorithm are presented, in order to gain a comprehensive trust value of the whole mobile commerce system. Thirdly, giving the fuzzy definition of nodes trust, the model can calculate local trust with fuzzy inference and obtain recommends from neighbor nodes. Lastly the proposed model is verified by an example.

The proposed fuzzy trust management model can help us manage the security and reliability of mobile commerce system regardless of differences in enterprise culture and business types.

Keywords Mobile E‐Commerce; Information Security; Trust Evaluation; Fuzzy Mathematics, Recommend

Introduction The trust and security problem is one the hardest problems in mobile commerce application, (Bhalaji, 2012; Chou, 2014; Hao et al, 2014; Abdel‐Halim et al., 2015). Since there are great uncertainty and subjective factors in mobile commerce system, the early researches mainly focused on mobile payment trust (Chou, 2014). It is showed that from the point of view of the original trust, the impacts of the two factors affected the performance and willingness of users (Chou, 2014). Trust benefits people who live in a risky and uncertain situation by providing means to decrease complexity, and it is the key to decision making, by which visualizing trust information can benefit users’ behavior and decisions (Suriavel et al., 2014; Wei et al., 2014).

On the one hand, these researches provide empirical evidences to illustrate the coexistence of integrity trust and integrity distrust. Generally, if an individual trusts the integrity of a service provider, the cause is more likely to keep its promise of providing genuine and personalized mobile services by a substantial investment in location detection technologies (Wang et al., 2013). On the other hand, if the individual distrusts the integrity of a service provider, it is the subsequence of this investment, where the service provider has a strong incentive to cheat in order to justify the investment (Suriavel et al., 2014; Yang et al., 2015).

However it is difficult to provide a theory framework of influence factors of consumersʹ trust for mobile commerce system, and there are several reasons as follows.

Firstly, due to lack of punishment to malicious recommendation from referee nodes, enormous recommendation nodes with subjective malicious tendency are remained in network and they are constantly seeking the next criminal opportunity. Bhalaji (2012) provide a dynamic trust based method to mitigate greyhole attack in mobile Ad Hoc networks, and Zhang et al. (2013) verified the effects of location personalization on integrity trust and integrity distrust in mobile merchants.

Secondly, the trust factors in mobile commerce system are very complex (Nilashi et al. 2015). For example, Liu et al.

(2010) discussed negotiation support framework for establishing bilateral trust in mobile commerce, and then Liu et al. (2013) further considered a cloud‐based trust establishment protocol towards mobile commerce. Therefore, how to make quantified evaluation on these trust factors in different mobile commerce businesses is very difficult.

Thirdly, the trust formation and transition mechanism of mobile commerce is unclear (Yang et al., 2015). Wang et al.

(2013) studied the transition of electronic word‐of‐mouth services from web to mobile context, and built a trust International Journal of Engineering Practical Research, Vol. 4 No. 2‐October 2015 115 2326‐5914/15/02 115‐08, © 2015 DEStech Publications, Inc. doi: 10.12783/ijepr.2015.0402.02

116 ZHENGYING CAI, JUN CHEN, YIJUN LUO

transfer perspective. Then Wang et al. (2014) tried to enable reputation and trust in privacy‐preserving mobile sensing. Some researchers designed different questionnaires on trust evaluation of mobile commerce, and made reliability analysis and validity analysis on the trust evaluation scale respectively.

Lastly it is very difficult to build a compatible trust evaluation method for mobile commerce system (Nilashi et al.

2011). Piao et al. (2010) put forward a mobile commerce trust model and its application for third party trust service platform, and Piao et al.(2012) further research on trust evaluating model for mobile commerce based on structural equation modeling. In fact, the enterprises and businesses bounded in the mobile commerce system are different from each other.

To measure these unclear factors and parameters, fuzzy mathematics is introduced, and is very popular in many economic and social areas. For example, Hao et al. (2014) applied fuzzy mathematics in mobile social network and proposed a Mobi Fuzzy Trust, which is verified as an efficient trust inference mechanism in mobile social networks.

Similarly, fuzzy mathematics can also be applied in mobile commerce system to observe the impact of trust information visualization on mobile application usage (Yan et al., 2013).

This paper proposes a dynamic recommendation trust evaluation model based on mobile e‐commerce environment. On the basis of the research questions above, we proposed a fuzzy trust measurement model for trust evaluation. Then the recommendations will be combined with local trust after getting synthesis weight by fuzzy inference information to obtain comprehensive trust of all nodes.

System Flow Diagram Generally, the trust value is also attributable to relationships within and between mobile commerce system or social groups, including all kinds of business partners, families, friends, companies, communities, organizations, nations, etc. The trust value is a popular approach to evaluate the dynamics interactions of inter‐group and intra‐group in mobile commerce system.

The system flow chart describes the whole trust formation and transition process in mobile commerce system. The initialized trust value includes recommended value, namely trust degree of recommendation, which will be compared with the benchmark value in this system. Trust value formation depends on observation and recommendation of the third parties such as service providers and users. The trust flow diagram is shown in Figure 1.

FIG.1 SYSTEM FLOW DIAGRAM

The whole process of trust formation and transition are as follows. Firstly, it is to assess the initialized trust value of local node of mobile commerce system by observing the change in the trust factor and combination with the fuzzy inference to be updated. Secondly, it is to assess the trust value of node by collecting recommendation information from other users and service providers. Finally, it is to assess recommendation from the local nodes by using fuzzy reasoning with its weights of recommendation trust value from comprehensive rights credibility, on the basis of which the recommended trust value in the synthesis assessment will form a comprehensive trust value of the system.

A Fuzzy Trust Measurement Method for Mobile E‐Commerce 117

Modern information technologies such as mobile commerce system, not only facilitated the trust formation and transition towards post‐information society, but also challenged traditional formation and transition on trust. In Figure 1, the trust was assessed by a distributed algorithm in the mobile commerce system, and the two adjacent nodes trust each other, where the assessment body is called as assessment node, while the objects are called evaluated nodes.

Fuzzy Trust Value In mobile commerce systems, a trusted component such as user or service provider has a set of characteristics which another component will rely on. In a business, if component a trusts b, this tells us that a violation in the properties of b will result in the correct operation of component a. Hence to observe the properties of b trusted by a, it will not correspond to the actual properties of B quantitatively or qualitatively. This means when the trust managers in the overall mobile commerce system need not to take these relation into account. Especially, trust should be placed to a certain context of the component trustworthiness, which is defined by a set of functional and non‐functional characteristics and how well it operates. Hence the trust of mobile commerce can be derived from the inner structure and outer environment of the system, namely it can be evaluated by appropriate method, including fuzzy mathematics.

In the mobile commerce networks, subjective evaluation of trust nodes mainly appears fuzzy, where trust classification is based on several attributes, instead of a simple binary logic. That is to say, a node could not be simply classified as a trusted or untrustworthy, a middle state could be taken into consideration. Hence, trust is not either‐or, but both‐and, that means a node not just belongs to a category of trust, which may belong to more than one trust category in varying degrees. To reflect this subjective fuzziness, the node trust value based on fuzzy set theory is determined as follows.

First of all, node trust is classified as several grades. Let x be the nodeʹs credibility, its domain is TX  d |d  [0,1] .

The trust classification nodes are described as five variables: ʺabsolutely incredibleʺ, ʺcompared unreliableʺ, ʺuncertainʺ, ʺless credibleʺ and ʺabsolutely credibleʺ. The corresponding fuzzy subset are X 1 , X 2 , X 3 , X 4 and X 5 , and the corresponding membership function are  X1 ,  X 2 ,  X 3 ,  X 4 and  X 5 . According to the trust classification of nodes, the trust value of the node is represented as a vector





TD  rD1 , rD2 , rD3 , rD4 , rD5 (1)

Where, (k = 1, 2, …, 5) is the membership degree of the node trust category Dk .

Evaluated nodes will be assessed on its own observations by comprehensive trust factors of multiple nodes to quantify the value of trust result, and be called as a local trust value Z BC , whose vector form is marked as





Z BC  ZCB1 , ZCB2 , ZCB3 , ZCB4 , ZCB5 (2)

Quantization process of local trust value is divided into two phases, initialization and update. In the process of quantization of local trust value, the observation of the trust factor is a subjective cognitive process, which is of strong fuzziness, where credibility trust factor with multiple values (observation that trust factor) is a comprehensive assessment of nodes when calculating weights are needed to determine a problem. Therefore, the model (2) uses fuzzy inference trust value of local node in calculation, not only embodying the trust quantifying subjective process, but also preventing the problem of weight uncertainty of the trust factor in calculating.

Formation and Transition of Trust Value Formation and transition of trust value is a process used in the mobile commerce system to determine which component should bear the formation of trust for users occurring to goods or services after the business has been completed, but before transition has occurred. Transition of trust value generally comes into play after the trust is formed, but before the downstream notes receive goods or service, something interaction will happen.

118 ZHENGYING CAI, JUN CHEN, YIJUN LUO

It is supposed that there are n trust factors: H1 , H 2 ,  , H n ( j = 1, 2, …, n) and they can be classified as m grades: K j1 , K j 2 ,  , K jm , thus the corresponding fuzzy subsets of its membership function are  kj1   i  ,  kj 2   i  ,  ,  kjm   i  ,

respectively, which represents the I‐type trust factor.

Thus, the local trust value is calculated as follows. ① Constructing fuzzy inference rules. According to rule knowledge and experience of trust reasoning, the fuzzy rules of node trust factor can be gotten by derived category, the total rules is set as W: Rule 1: if H1 is Kj1, and H2 is Kj2, and … and Hn is Kjn, then X is X1; Rule 2: if H1 is Kj1, and H2 is Kj2, and … and Hn is Kjn, then X is X1; Rule W: if H1 is Kj1, and H2 is Kj2, and … and Hn is Kjn, then X is X5; ②Extracting fuzzy implication relationship S K  X . At first, it is to strike implication relations under a single fuzzy rule S  = (  =1,2,…,W) with formula

 S    k1   1    k2   2    km  m    x  x 

(3)

 Where  k1  1  ,  k2   2  ,…,  km   m  are the first fuzzy rules under the membership functional subset. Secondly all

the fuzzy comprehensive implication relations are under the rules, you can get this rule with multiple relationships as follows.





  m   x x (4) SKX  W1 K1 1   K2  2   Km

③ Reasoning the local credibility of every nodes. Using the actual value of the trust factor 1 , 2 ,  , m and synthetic relationship SK  X , get local credibility fuzzy output node, such as formula (5)   X  X   W 1   K1  1    K 2   2      Km  m   x  x   (5)

Then using the center of gravity degasification method, the credibility of the local nodes can be obtained, as the formula (6) X   COG    X  X   XdX /   X  X   XdX (6)

④ Calculating the trust vector of local nodes. The use of node‐local credibility, combined membership function of each nodeʹs local trust value, the trust category calculated as formula (7)





Z BC   X 1 ( X  ),  X 2 ( X  ),  X 3 ( X  ),...,  X 5 ( X  ) (7)

Therefore, the proposed model is designed as follows. Firstly, it is to assess the local nodes’ trust value, then to combine it with the fuzzy inference to update it information by observing the change in the trust factors. Secondly, it is to assess node recommendation gathered from other nodes to assess the trust value of nodes. Finally, the assessed node is recommended by its local nodes by fuzzy reasoning and comprehensive right credibility from its recommendation trust value weights, where recommended trust value in the synthesis of the assessment can produce the comprehensive trust value of the whole system.

Trust management Process Step 1. Initialization of the Local Trust Value After the establishment of the neighbor relationship of nodes, all nodes can be reviewed to assess the initialize local trust value of the other nodes. In case of lacking of prior information, to assess node can not be trusted to determine the assessment condition of nodes and will be evaluated by local trust values as a uncertain component of the assignment process. Let the other assignment be 0, that is Z BC   0, 0,1, 0, 0  , and the initial values of 1 , 2 ,

A Fuzzy Trust Measurement Method for Mobile E‐Commerce 119

 , m are all 0.

Step 2. Recommendation of Trust Value





Recommendation of trust values is indicated by vector form EIT  ETI 1 , ETI 2 , ETI 3 , ETI 4 , ETI 5 . In order to cut energy waste in communication and prevent the recycle loop of the trust, it is necessary to limit neighboring nodes without illegal pass iteration. The recommendation trust value provided by interrelated node is referred as a recommendation node. As shown in Figure 1, the recommendation nodes would automatically send the value of the local trust of the nodes that is assessed as the recommendation value to the evaluated nodes after receiving the requirement of evaluation nodes, and its local trust value of the node can be evaluated as a recommendation trust value for the node trust assessment. To sum up, only nodes who are the neighbor with both evaluating nodes and being evaluated nodes by others can be able to provide recommendation trust value.

Nodes in mobile commerce system are usually deployed in densely. However, due to the randomness of the distribution of nodes, there is some special circumstance that nodes might have no recommendations. In this case, the assessment directly from the neighboring node is evaluated as an integrated trustable node.

Step 3. Integration of Trust Values Evaluating nodes will collect recommendation trust values and synthesis them to an integrated trust value of the local trust measurement, in order to improve the accuracy and robustness of trust assessment. Therefore the comprehensive trust value’s vector form is marked as: ECT   ETC1 , ETC2 , ETC3 , ETC4 , ETC5  . Node’s trust value is decided by two‐step process, overall weight computing and synthesis trust value.

Simulated Experiment and Analysis Based on the proposed theory, a research questionnaire is designed to collect appropriate indexes including quality of goods, quality of service, user satisfaction, and complaint. Referred to similar questionnaires used in prior research (Chou, 2014; Hao et al.,2014; Liu et al., 2010), weighing factors are added to each relevant subject and are revised in accordance with the characteristics of the mobile commerce system. Within the discretion of the academics with inspected content, relevant personnel factors are considered so that the simulated experiment can meet the requirements of the questionnaire contents. The experimental data are tested for reliability and validity using Mamdani fuzzy rule.

The mobile commerce system has two inputs: the user outside of the system and the service providers in the system. For both inputs, there are five fuzzy values defined: negative big (NB), negative small (NS), Zero (Z), positive small (PS), positive big (PB). For the output, the trust, exchanged with the user and service provides, seven fuzzy values are defined: negative big (NB), negative medium (NM), negative small (NS), Zero (Z), positive small (PS), positive medium (PM), positive big (PB). A positive trust transition flows when the business is operating. The membership functions of output and input are shown in Fig. 2.

FIG. 2 TRIANGLE RULE OF MAMDANI RULE

In this experiment, the transfer function of trust is focused on between the mobile commerce centers and the end users, together with the processing inventory or half‐ products between the center facilities and the user demand.

120 ZHENGYING CAI, JUN CHEN, YIJUN LUO

Especially in mobile commerce system, it is assumed that the supply rate of products from a upstream nodes is often imprecise, therefore the quantity can be also described as a fuzzy number for the simulation. The fuzzy Mamdani rule consists of 25 fuzzy logic rules, giving the trust output according to the two inputs of user and service provider as presented in Table 1, where there are five input language variables and seven output language variables.

TABLE 1 FUZZY RULE

Input Output

NB

NS

Z

PS

PB

NB

PB

PM

PM

PS

Z

NM

PB

PM

PS

PS

NS

NS

PM

PM

PS

Z

NS

Z

PM

PS

Z

NS

NM

PS

PS

Z

NS

NS

NM

PM

PS

NS

NM

NM

NB

PB

Z

NS

NM

NM

NB

For the imprecision of the trust formation and transition, the whole mechanism is described by fuzzy terms like big and small, which makes it possible to illustrate the complex transformation process of trust. Whereas the traditional crisp mathematics is difficult to interpret and develop reliable trust formation behavior. In addition, the amount of trust information exchanged between different nodes need to be estimated but is not clearly defined in the fuzzy rules. The rule results in the fuzzy view are shown on Fig. 3.

FIG. 3 FUZZY RULES

The simulated experiment is based on MATLAB 7.0, the initial scene is set as follows, it is assumed that the communication radius of each node is 25 m, there are 100 nodes randomly dispenser in a 100 m × 100 m detection area, and 10% of the nodes are malicious nodes, packet loss rate and packet tampering rates are set as 75% to 100%.

Under the collusion and strategic attack mode, the updating cycle of local trust value is assumed as 0.1 s.

FIG.4 THE FOUND PROPORTION OF MALICIOUS NODES BY THE PROPOSED MODEL AND TRADITIONAL MODEL

A Fuzzy Trust Measurement Method for Mobile E‐Commerce 121

The simulation result with traditional crisp trust measurement is shown in Fig. 4. The absolutely credible and less credible nodes are designated as malicious nodes in our experiments. We test the validity of fuzzy trust model to improve network security by observing the discovered proportion of malicious nodes, and compare with the traditional model. Fig. 4 shows that the two found proportions of malicious nodes are different, and under the fuzzy trust model the proportion of malicious nodes was found rise rapidly in the initialization phase of the network, and then maintained at a high level of 0.8.

The reason is that the proposed model defined node trust value by fuzzy value and quantified fuzzy inference to ensure the accuracy of trust evaluation, so the inclusive analysis of the weights and the robustness of trust assessment are enhanced. The proposed model can help us manage the security and trust of mobile commerce system regardless of differences in enterprise culture and business types.

Also, the model assumes that there is always enough supply of qualified products to satisfy the user demand and there is enough inventory space for unfinished and finished products, so the inventory trust is always higher because of higher availability at the inventory facilities. Furthermore, inventory costs at the commerce centers and service providers are not considered in our model, in fact, they are inevitable. For example, if the supply of service providers is higher than the user demand, the trust value may be higher, but the inventory centers have to bear more inventory costs.

Conclusion This paper presents a trust evaluation model based on fuzzy mathematics which gives the formal definition of the trust value of nodes by quantified tool. It can quantify the value of a local trust with the method of fuzzy reasoning and obtain recommendation information from its’ neighbors, combining it with local trust after getting synthesis weights with fuzzy inference. Finally comprehensive trust can be obtained by trustable nodes. However, the proposed model may be revised in next work, in response to existed constraints including inventory capacity, cost, management level, and other constraints. In future research, it is planned to extend the model to consist of more roles, multiple products or services and multiple periods, etc.

ACKNOWLEDGEMENTS

This research was supported by the National Natural Science Foundation of China (No. 71471102), and Science and Technology Research Program of Hubei Provincial Department of Education in China (Grant No. D20101203).

REFERENCES

[1]

Abdel‐Halim, Islam Tharwat; Fahmy, Hossam Mahmoud Ahmed; Bahaa‐Eldin, Ayman Mohammad. Agent‐based trusted on‐demand routing protocol for mobile Ad‐Hoc networks[J]. Wireless Networks, 2015, 21(2): 467‐483.

[2]

Bhalaji, N; Shanmugam, A. Dynamic trust based method to mitigate greyhole attack in mobile AdHoc networks [C].

International Conference on Communication Technology and System Design, 2012, 30: 881‐888.

[3]

Chou, Tao. An empirical examination of initial trust in mobile payment [J]. Wireless Personal Communications, 2014, 77(2): 1519‐1531.

[4]

Hao, Fei; Min, Geyong; Lin, Man; et al. MobiFuzzyTrust: An efficient fuzzy trust inference mechanism in mobile social networks[J]. IEEE Transactions on Parallel and Distributed Systems, 2014, 25(11): 2944‐2955.

[5]

Liu, bailing; Wang, Weijun; Wang, Jianzong. A negotiation support framework for establishing bilateral trust in mobile commerce [J]. Information‐An International Interdisciplinary Journal, 2010, 15(9): 3841‐3847.

[6]

Liu, bailing; Zheng, D; Shi, J; Zhang, L; Cloud‐based trust establishment protocol towards mobile commerce [C].

International Conference on Computer, Networks and Communication Engineering (ICCNCE), 2013: 700‐702.

[7]

Nilashi, M., Ibrahim, O., Mirabi, V. R., Ebrahimi, L., & Zare, M.. The role of Security, Design and Content factors on customer trust in mobile commerce[J]. Journal of Retailing and Consumer Services, 2015, 26, 57‐69.

[8]

Nilashi, M., Bagherifard, K., Ibrahim, O., Janahmadi, N., & Barisami, M. An application expert system for evaluating effective factors on trust in B2C[J]. Engineering, 2011, 3(11), 1063.

122 ZHENGYING CAI, JUN CHEN, YIJUN LUO

[9]

Piao, Chunhui; Wang, Shuzhen; Wen, Jie. Mobile commerce trust model and its application for third party trust service platform [C]. IEEE 14th International Conference on Commerce and Enterprise Computing (CEC), 2010:120‐125.

[10] Piao, Chunhui; Wang, Shuzhen; Yang, Fengtao. Research on Trust Evaluating Model for Mobile Commerce Based on Structural Equation Modeling [C]. 9th IEEE International Conference on e‐Business Engineering (ICEBE) / 8th SOAIC / 6th EM2I / 6th SOKMBI / 4th ASOC, 2012: 25‐32.

[11] Suriavel Rao R S, Malathi P. A survey on Resource Optimization Techniques of MC‐CDMA System used in 4G Wireless Mobile Communication systems[J]. International Journal of Engineering Practical Research (IJEPR), 2014, 3(3): 59‐65.

[12] Wang, Xinlei (Oscar); Cheng, Wei; Mohapatra, Prasant; et al. Enabling reputation and trust in privacy‐preserving mobile sensing[J]. IEEE Transactions on Mobile Computing, 2014, 13(12): 2777‐2790.

[13] Wang, Nan; Shen, Xiao‐Liang; Sun, Yongqiang. Transition of electronic word‐of‐mouth services from web to mobile context: A trust transfer perspective [J]. Decision Support Systems, 2013, 54(3): 1394‐1403.

[14] Wei, Zhexiong; Tang, Helen; Yu, F. Richard; et al. Security enhancements for mobile ad hoc networks with trust management using uncertain reasoning[J]. IEEE Transactions on Vehicular Technology, 2014, 63(9): 4647‐4658.

[15] Yan, Zheng; Liu, Conghui; Niemi, Valtteri. Exploring The impact of trust information visualization on mobile application usage[J]. Personal and Ubiquitous Computing, 2013, 17(6): 1295‐1313.

[16] Yang, Shuiqing; Chen, Yuangao; Wei, June. Understanding consumersʹ web‐mobile shopping extension behavior: a trust transfer perspective[J]. Journal of Computer Information Systems, 2015, 55(2): 78‐87.

[17] Zhang, Ruidong; Chen, Jim Q; Lee, Ca Jaejung. The effects of location personalization on integrity trust and integrity distrust in mobile merchants [J]. Mobile Commerce and Consumer Privacy Concerns, 2013, 53(4): 31‐38.