Show Summary Details
More options …

# Open Physics

### formerly Central European Journal of Physics

Editor-in-Chief: Seidel, Sally

Managing Editor: Lesna-Szreter, Paulina

IMPACT FACTOR 2018: 1.005

CiteScore 2018: 1.01

SCImago Journal Rank (SJR) 2018: 0.237
Source Normalized Impact per Paper (SNIP) 2018: 0.541

ICV 2017: 162.45

Open Access
Online
ISSN
2391-5471
See all formats and pricing
More options …
Volume 15, Issue 1

# Analysis on trust influencing factors and trust model from multiple perspectives of online Auction

Wang Yu
Published Online: 2017-10-31 | DOI: https://doi.org/10.1515/phys-2017-0071

## Abstract

Current reputation models lack the research on online auction trading completely so they cannot entirely reflect the reputation status of users and may cause problems on operability. To evaluate the user trust in online auction correctly, a trust computing model based on multiple influencing factors is established. It aims at overcoming the efficiency of current trust computing methods and the limitations of traditional theoretical trust models. The improved model comprehensively considers the trust degree evaluation factors of three types of participants according to different participation modes of online auctioneers, to improve the accuracy, effectiveness and robustness of the trust degree. The experiments test the efficiency and the performance of our model under different scale of malicious user, under environment like eBay and Sporas model. The experimental results analysis show the model proposed in this paper makes up the deficiency of existing model and it also has better feasibility.

Keywords: trust model; reputation; online auction; Sporas; weight

PACS: 42.30.Sy

## 1 Introduction

E-commerce effectively reduces the expenditure of social costs and provides a strong guarantee for the transformation and transformation of traditional industries, to achieve overall economic progress, and it has more obvious positive function on the social development [1,2,3,4,5]. As a novel market transaction pattern, online auction greatly expanded the scope of traditional auction market [6], which also reduces the market transaction costs and improves the efficiency of market transactions. But the issues of fraud and breach of contract still exist universally in online auction and they are becoming increasingly acute. Most of current auction sites are absent in perfect trust mechanism to reduce or avoid similar problems. Therefore, the research on reputation, the reference index between both sides, to well know the conditions of credibility of each other, shows particularly important. The design of scientific and reasonable reputation feedback system has great significance on establishing perfect online auction trust guarantee mechanism, to promote the healthy development of online auction.

The concept of trust idea is early proposed by B1aze et al. [7], which is an approach to illustrate and explain the safety strategy, certificate and trust relation uniformly. It can cross a number of management domains to achieve accurate identifications. Based on this theory, people propose various subjective trust models such as probability models of simple sum or average value, and discrete trust model, flow model, etc. Large online auction websites like eBay and Yahoo adopt a simple accumulation or mean method to compute the degree of trust. It is found that the reputation computation and structure of current trust models are not suitable to the reputation evaluation for transaction of both sides and they lack descriptions of characteristics of online auction transactions [8, 9]. The client behaviors are unpredictable and malicious behaviors on trust computation cannot be solved. Simultaneously, the process of reliability computation on recommendation message from reputation is relatively simple, neglecting the evaluation problems such as different positions of participants in online auction. Thus, a scientific and effective online auction trust model is needed urgently to simulate the dynamic transaction process of online auction objectively and study the trust management and trust computing methods, to ensure a stable and robust development environment for the online auctions.

This article adopts a combination of theoretical research and empirical research to perform modeling on online auction reputation evaluation system and efficiency analysis. First it investigates the features of online auctions and the influencing factors obstructing the development. The comprehensive factors of three types of participants are studied such as credibility, auction evaluation, and behavior analysis. Then the deficiency of current reputation feedback system and these trust influencing factors are introduced to the model. Sporas model [10] is improved to form an optimized reputation feedback evaluation model, through the evaluation of both sides. The empirical research results indicate that the multiple perspective trust model greatly reduce the influence of interference factors in trust evaluation, which improves the accuracy, validity and soundness of the trust degree of model.

The remainder of this paper is organized as follows. In Section 2, a description of the trust model and reputation evaluation system is given in detail. In Section 3, a multiple perspectives based trust model is proposed and the key parameters influencing the value are discussed in detail. In Section 4, we use simulation analysis to verify the effectiveness of our model and conclusions are presented in Section 5.

## 2.1 Concept model of trust

The definition of trust in online auction is a continuation of the concept of traditional trust [11]. Because of the uncertainty and risk of exchange dependent environment, the trust between traders directly affects the success or failure of online auction trading. Therefore, the research of trust focuses on trading body and electronic market. The main body of transaction is a trader including personal characteristics, trust and other related factors. E-commerce web site, including network, technology and other related factors, which often plays an important role in establishing trust model, and it can be analyzed from two perspectives: subjective factors and objective factors.

The basic frame work of online reputation evaluation system is depicted as Figure 1. Assuming Ai(i = 1, 2, …, m) is individual that has transaction with B in a community. VB denotes the credit rating from Ai to B, The credit evaluation from A to B is decided by evaluation ei from all individuals Ai to B. The online reputation evaluation system collects all the historical credit marks by portal site and summarizes the feedback marks from A and B, according the reputation information. Then through the portal website, the final reputation information is published to all potential traders. Besides the evaluation from all other individuals to certain individual will affect its credit degree, the accuracy of its evaluation to others is seen as one of the honesty reference of itself.

Figure 1

The structure of an online reputation evaluation system

## 2.2 Sporas model

In C2C E-commerce transaction platform, once a trade is completed, the centralized online reputation feedback system will automatically gather the feedback mark information of both sides and computes bilateral reputation value based on relative transaction factors. The fundamental purpose of trust degree model is reflecting the true credibility of both sides through reputation degree, to provide a referable signal for potential participants and effectively reduce the information collecting cost under E-commerce [12].

Sporas model is set to follow the principles below:

1. New registered users of trading platform have the lowest initial reputation value, and the growth of value of its reputation is accumulated by participating in the transaction.

2. When the reputation value decreases it will not be lower than initial reputation value. Because when the user’s reputation is lower than initial reputation caused by credit, he tends to give up current identity account normally to reregister as a new user to attract potential transaction objects.

3. After each transaction both sides will offer mutual evaluation and the online reputation feedback system of trading platform will update the credit value of both sides. The model is depicted as $Ri=Ri−1+1θ(Ri−1)Riother(Wi−Ei)$(1)

Ri and Ri−1 denote the trust degree of user at hour i and i−1; Wi denotes the user reputation feedback mark at hour i; $\begin{array}{}{R}_{i}^{other}\end{array}$ denotes the trust degree of raters posting Wi. Though Sporas model improves some necessary problems in trust computation, it does not consider the online transaction factors such as transaction number, transaction price, scoring time etc. That is to say, there still exists deficiency.

## 3.1 Participation classification and its influencing factors

For each transaction participation entity in online auction, since the perspective of itself is different, the participating mode in auction is also different. When evaluating the trust degree of participants, the trust model needs to investigate trust degree of different participants from various perspectives [13]. Affecting by different factors, the participants will generate different trust degree.

Figure 2

Participants of online auction

The bidder means a party that wants to obtain resources in an auction; The tender is the he party owing the resources in auction activities and formulating the rules; The third party refers to the network auction activities, supervisors and managers relying on the auction website. In traditional auction activities, as long as there is an auction process with bidders and tenders to participate in, the auction is a complete auction. However, online auction not only involves bidders and tenders, but involves the third party agencies or auction sites. The tender and the bidder must rely on the process of online auction, and be subject to their management, supervision and restriction.

In the trust model from multiple perspectives to participate the entity evaluation trust degree, it often suffers influence from the same or different factors. In detailed evaluation of trust degree, the parameters in the following table are needed to be investigated:

From Table 1 we can see, that recent reputation, reputation feedback score, reputation of scoring user, time weight and transaction value have certain influence on the reputation. Therefore, according to analysis on the key influencing factors, we improve Sporas trust model and propose a reputation model based on multiple perspectives in C2C online auction.

Table 1

Parameters table of transaction entity evaluation

## 3.2 Reputation model description

In given time t, the reputation computation model of user u is: $Rt(u)=αRt−1(u)+β∑x∈R(u)λ[p(x,u)]⋅Cr[Rt−1(u)]δ(Δt)f¯(x,u),N(u)≠0Rt−1(u),N(u)≠0$(2)

Rt(u) is the reputation degree of u at time t; N(u) is the set of transaction partners, denoting the set of trading users who has transaction with u in the time domain [t−1,t]; α and β denote the reputation feedback scores posted by trading partners xN(u) of user u and recent reputation degree respectively. λ [p(x,u)] is transaction amount weight function; Δ t denotes the duration time of auction; Cr[Rt−1(u)] is weight of reputation feedback score; f(x,u) is the average reputation score of u when the auction is over. It can be depicted as $f¯(x,u)=∑i=1|C|ωcifci(x,u)∑i=1|C|ωcici$(3)

where |C| where denotes the basis of the set of reputation key factors.

This model includes two parts: the first is the reputation degree of user when time window is t−1. It aims at emphasizing the influence on current reputation of recent reputation status. The weight mainly considers factors such as the reputation degree of reputation feedback scoring person, transaction value and time discounting. For a simple discussion we can just simplify the model when α = β = 1. Then the simplified form of model is: $Rt(u)=Rt−1(u)+∑x∈R(u)λ[p(x,u)]⋅Cr[Rt−1(u)]ρ(tx,t)f¯(x,u),N(u)≠0Rt−1(u),N(u)≠0$(4)

Due to the simplified model we know the value of Cr[Rt−1(u)] needs a further research to get a perfect trust model.

## 3.3.1 Weight of transaction amount

This paper uses logical growth curve model in narrow sense to compute the transaction amount weight. First we provide a function describing the logic growth curve model, commonly known as “S curve” [14]. It is widely used in economic application and it is an important prediction model with the usual form as $yt=K1+ae−bt$(5)

This curve performs a translation with α unit along the horizontal axis to get the following function $y=11+e−(x+α)$(6)

We directly adopt the online shopping value v(y,x) as transaction amount. Its value range is (0,+ ∞), which may be less than 1 or more than 1. For convenient analysis of reputation score, v(y,x) is mapped as λ [p(x,u)], to let λ [p(x,u)] ∈ (0,1). Since the average transaction price in our country is about 200 according to relative literature [15], we get the detailed transaction amount weight function as $λ[p(x,u)]=11+e−3100p(x,u)+6$(7)

## 3.3.2 Weight of scoring time

The scoring time function between weight function δ(Δt) and Δt is δ(Δt) = f(Δt). The value of δ(Δt) is between 0 and 1 and the function of f is mapping Δt, which has the following characteristics: when t−1 approaches to t, if Δt is smaller, the weight of reputation score Rt−1(x) is higher. According to this characteristics, we set φΔt. If φ = 0.1 we have $f(Δt)=0.1Δt$(8)

When Δt = 1, f(Δt) = 0.1. It indicates the interval is one day between two completed transactions, and the scoring time weight of historical reputation is 0.1, which is close to 0 and reasonable. Then we stretch this curve γ times along the horizontal axis to get $δ(Δt=1)=f(Δt)=0.1Δtγ$(9)

## 3.3.3 Weight of reputation feedback score

On the basis of collaborative filtering algorithm [16], we propose an improved scheme using the correlation among item attributes and compute the final prediction score according to the influence among different items. Usually we use Person correlation degree to compute the correlation among items. But this method is subjective and is easy to be influenced by users, so it cannot reflect the correlation among items and will cause inaccurate recommendation quality. Therefore, we propose to compute the item correlation based on attributes matrix. There is an item attributes matrix Attr = {attrij}. attrij denotes whether the ith item has jth attribute, or its value is any integer of 0 or 1. Then the attributes matrix is: $attrij=001…1010…1110…0⋮⋮⋮⋱⋮100…1$

The computation equation of item similarity is $rel(i,ip)=∑j=1kij&ipj|Attr(i)∪Attr(ip)|,∑j=1kij&ipj<ω$(10)

|Attr(i)∪ Attr(ip)| denotes the number of attribute union set of items waiting to be scored with other items; $\begin{array}{}\sum _{j=1}^{k}{i}_{j}\phantom{\rule{thinmathspace}{0ex}}\phantom{\rule{thinmathspace}{0ex}}\mathrm{&}\phantom{\rule{thinmathspace}{0ex}}\phantom{\rule{thinmathspace}{0ex}}{i}_{pj}\end{array}$ denotes the number of common attributes of items waiting to be scored with other items. ω is the maximum value of common attributes, that is, it denotes the number of public attributes between 2 items. When it is beyond the maximum value it gets absolutely similar and we have $rel(i,ip)=1,∑j=1kij&ipj≥ω$(11)

When the item correlation is introduced, the similarity of users can be described as follows: $sim(ua,ub)ip=∑i∈Uab(uai−u¯a)(ubi−u¯b)∗rel(i,ip)∑i∈Uab(uai−u¯a)2∗∑i∈Uab(ubi−u¯b)2$(12)

where sim(ua,ub)ip is the similarity of user a and b on item iP that is to be predicted. Finally it is substituted to the original filtering equation to get the value of Cr[Rt−1(u)].

## 4 Experimental analysis

We use Matlab to establish a C2C simulation platform for E-commerce. Assuming the users do not know each other and the online goods displayed of each online user can be seen by any other users. We set an authenticity parameter T to denote the success or failure of user. If T = 1, the trading succeeds; otherwise, the trading fails. Each user in the model saves a list of trust value, a list of recommendation and an authenticity parameter. The parameter value is assigned as 0 or 1 in the experiments. Another parameter of threshold is set to denote the psychologic limit of user. If trading cost is beyond this value, the user chooses to give up. In addition, we classify the users to different classes of persons. We can also set the fraud rate in the whole network and update the trust list and recommendation list when each transaction is completed.

Each trading process is triggered by a random user in the experiment. The person with different classes at the initial stage may be honest traders or trickers. The users search for the trading users whose trust value is higher than the threshold by querying his own trust value table. If wanted user exists, the transaction is performed directly. Otherwise, he sends querying information to other users and computes the comprehensive trust value of users who have higher trust value for a transaction.

The effectiveness of the model in this paper is verified by the similarity of experimental subject and actual data. The similarity and dissimilarity among the objects are computed by the distance among them. The most frequently used method is Euclidean distance. The two-dimensional equation is $dij=|xi1−xj1|2+|xi2−xj2|2+…+|xim−xjm|2$(13)

Illustrated by the bidders, we compute the distance between trust degree and simulated trust degree, as well as the distance between simulated distance and trust value acquired by the improved model proposed in this article. When the Euclidean distance is small, it indicates that the corresponding trust value is closer to simulated trust degree.

We suppose the distance between trust degree and simulated trust degree is d1, and the distance between simulated distance and trust value acquired by the improved model is d2. After simulation we can get d1 = 0.59 and d2 = 0.75, as shown in Figure 3. Since d1 > d2, the results of trust mode is better than simulated trust degree, so our model is effective.

Figure 3

The trust comparison with Euclidean distance

To test the effect of Fraud behaviors in trading, we compare our model with common Sporas model. The trust judgment standard is: whether the nodes in the electronic commerce network have the historical interaction is taken as the only condition of trust [17]. So we design the following experiment: when the nodes in network perform a period of honest transaction, some of them may make some bad behaviors for their own interests. After the observation of the change of trust value, depicted in Figure 4, we can draw the following conclusions:

Figure 4

Influence on reputation by fraud behaviors

1. The response of our mode on fraud behaviors of malicious nodes is quicker and the range of trust value is larger. When there are dishonest transactions from supplier nodes, the trust value will drop quickly. Then it shows the model can effectively find the malicious behaviors.

2. With the increase of the rate of malicious recommendation, compared to traditional Sporas model, our model has better tolerance ability. When the number of malicious nodes increases continuously, the success rate of general trust model gets a quick plunge. However, the trading success rate of the improved model still keeps 50 percent or so when the malicious recommendation nodes accounted for 50 percent in the network. It indicates the error-tolerant rate is higher for our model.

It can be analyzed, based on the above content, the trust model in online auction based on multiple perspectives is a more specific and finesorted model, compared to other computational models, which are verified to adapt to different trust degree of traders in online auction. The designing idea of multiple perspectives improves the evaluation accuracy and robustness of trust degree. The appropriate selection of parameters also improves the fraud prevention ability of the model.

## 5 Conclusion

Towards the deficiency of current trust computation methods and limitation of traditional trust model of online reputation system, this article provides a multiple perspectives based analysis. It summarizes various items of important trust factors influencing online auction trust and establishes multiple influencing factors based on trust model. The experimental results verify the effectiveness and feasibility of the improved model. There is a large amount of computation and computational complexity in the existing trust model. For the problem of large error in the calculation of trust, we can try to use some intelligent optimization methods to optimize the structure and trust computation. Therefore it will be the future research work to increase the robustness of presented model

## Acknowledgement

Research on the allocation mechanism of carbon emission right based on the different development stages in China, The National Social Science Funds Project (14BJL101).

## References

• [1]

Pan X., Ju X.-F., A research on development of Chinese B2C E-commerce trust, Proceedings of International Conference on E-Business and Information System Security, 2009, 167-181. Google Scholar

• [2]

Malaga R. A., Web-based reputation management system, problems and suggested solutions, Electronic Commerce Research, 2001, 1, 403-417.

• [3]

Dehning B., Richardson V.J., Urbaczewski A., Reexamining the value relevance of E-Commerce initiatives, Journal of Management Information Systems, 2004, 21(1), 55-82.

• [4]

Sarkar A., E-Commerce adoption and implementation in automobile industry, A case study, World Academy of Science, Engineering and Technology, 2009, 58, 587-594 Google Scholar

• [5]

Shi Y., Lin W., Yang J.-Z., An empirical study on urban ecommerce competitiveness of China. International Journal of Mobile Communications, 2016, 14(4), 328-347

• [6]

Lao F., W. X.-Y. Study on the E-commerce platform based on artificial intelligence. RISTI - Revista Iberica de Sistemas e Tecnologias de Informacao, 2015, 16, 173-188 Google Scholar

• [7]

Blaze M., Feigenbaum J., Lacy J., Decentralized Trust management, Proc. Of the 17th symposiumon Security and Privacy. IEEE Computer Society Press, Los Atamitos, 1996. Google Scholar

• [8]

Gregg D.G., Walczak S., Auction Advisor, An agent-based onlineauction decision support system. Decision Support Systems, 2006, 41(2), 449-471.

• [9]

D’Andrea A.E., Dologite D.G., Mockler R.J. et al. eBay, Inc., The online auction industry. Annals of Cases on Information Technology, 2004, 6, 41-57. Google Scholar

• [10]

Zacharia G., A’Moukas P.M., Collaborative reputation mechanisms in electronic marketplaces, Decision Support Systems, 2000, 29(4), 371-388.

• [11]

Peng G., Liu F., Chen J., The impacts of reputation on online auction-an analysis of Taobao. Journal of Computational Information Systems, 2009, 5(2), 781-786. Google Scholar

• [12]

Lin Z.-X., Li J., The online auction market in China - A comparative study between Taobao and eBay, Proc. of ACM International Conference Proceeding Series, 2005, 123-129. Google Scholar

• [13]

Bockstedt J., Goh K., Seller strategies for differentiation in highly competitive online auction markets, Journal of Management Information Systems, 2011, 28(3), 235-267

• [14]

Malinen S., Ojala J., Perceptions of trust between online auction consumers. International Journal of Web Portals, 2011, 3(4),15-26

• [15]

iResearch, Research Report on online auction in China of 2014 [EB/OL], http://www.iresearch.com.cn

• [16]

Cohen W.W., Fan W., Web-collaborative filtering, recommending music by crawling the Web. Computer Networks, 2000, 33(1), 685-698

• [17]

Popa D.M., Nica C.G.B., Trust and reputation systems in two online auction and shopping websites from Romania and the Netherlands. Proc. of DAAAM International Symposiumon Intelligent Manufacturing and Automation, 2012, 601-604. Google Scholar

Accepted: 2017-02-23

Published Online: 2017-10-31

Citation Information: Open Physics, Volume 15, Issue 1, Pages 613–619, ISSN (Online) 2391-5471,

Export Citation