research article a novel iterative and dynamic trust...

Research ArticleA Novel Iterative and Dynamic Trust Computing Model forLarge Scaled P2P Networks

Zhenhua Tan1 Xingwei Wang12 and Xueyi Wang1

1Software College Northeastern University Shenyang 110819 China2College of Information Science and Engineering Northeastern University Shenyang 110819 China

Correspondence should be addressed to Zhenhua Tan tanzhmailneueducn

Received 30 October 2015 Accepted 11 January 2016

Academic Editor Seung Yang

Copyright copy 2016 Zhenhua Tan et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Trust management has been emerging as an essential complementary part to security mechanisms of P2P systems andtrustworthiness is one of the most important concepts driving decision making and establishing reliable relationships Collusionattack is a main challenge to distributed P2P trust model Large scaled P2P systems have typical features such as large scaleddata with rapid speed and this paper presented an iterative and dynamic trust computation model named IDTrust (Iterative andDynamic Trust model) according to these properties First of all a three-layered distributed trust communication architecturewas presented in IDTrust so as to separate evidence collector and trust decision from P2P service Then an iterative and dynamictrust computation method was presented to improve efficiency where only latest evidences were enrolled during one iterativecomputationOn the basis of these direct trustmodel indirect trustmodel and global trustmodel were presentedwith both explicitand implicit evidences We consider multifactors in IDTrust model according to different malicious behaviors such as similaritysuccessful transaction rate and time decay factors Simulations and analysis proved the rightness and efficiency of IDTrust againstattacks with quick respond and sensitiveness during trust decision

1 Introduction

Large scaled P2P network is developed rapidly in recentyears because of its openness anonymity and being self-orga-nized P2P networking traffics always occupy Internet andP2P architecture is an important part of future Internet [1]Various P2P applications such as Mobile P2P P2P lending(eg Zopacom) and P2P e-commerce (eg eBay) have beenwelcomed by people and demonstrate high-capacity varietyand rapid response [2ndash4]

Trust management has been emerging as an essentialcomplementary part to security mechanisms of P2P systemsand trustworthiness is one of the most important conceptsdriving decision making and establishing reliable relation-ships A well-defined trust model can provide meaningfuldecision support and help customers reduce possible risksduring an Internet transaction Like trust and reputation insocial networks trust evaluation in P2P systems is based ontransaction histories

Threat and attack always exist in P2P networks Somecommon attacks against trust management remain in P2Pnetwork such as Sybil Attack Newcomer Attack BetrayalAttack Inconsistency Attack Bad-MouthingBallot StuffingAttack and Collusion Attack [5ndash8] According to the behav-ioral characteristics of malicious nodes they can be dividedinto individual and collusion malicious These maliciousbehaviors destroy usersrsquo trust in P2P systems Individualmalicious nodes tend to act alone and make fault evaluationto any other nodes while collusion malicious nodes act as ateam causing more serious consequence

In large scaled P2P networks how to quickly calculate themass trust evidences is a challenge to the efficiency of trustcomputing and how to defend against collusion maliciousbehaviors is also a big challenge hindering the effectiveness oftrust managementThemain objective of this paper is to pro-pose an iterative and dynamic trust computingmodel namedIDTrust for short to effectively evaluate nodes trust degreeagainst both individual and collusive malicious behaviors in

Hindawi Publishing CorporationMobile Information SystemsVolume 2016 Article ID 3610157 12 pageshttpdxdoiorg10115520163610157

2 Mobile Information Systems

large scaled P2P systems First of all explicit and implicitevidences are defined in IDTrust and then direct trustindirect trust global trust and decision trust are modeled inmathematical expressions based on evidencesThis paper hasthe following innovative features

(1) How to obtain large number of trust evidences andmake metric decision iteratively and dynamically isthe key in trust computing within an environmentof large scale P2P Internet As a result we proposea kind of computing architecture with hierarchicaldistribution trust which layers these three parts inde-pendently evidence collection trust computing andP2P communication leading to a three-layered P2Ptopology and enhancing the distribution computingof P2P commercial communication

(2) Traditionally each trust computing involves all thetrust evidence which contradicts the characteristicof P2P rapid and large scaled data Therefore in thispaper we propose an iterative and dynamic trustcomputing model Within this model trust comput-ing is iterative and only requires the newest trust evi-dence The performance of trust computing is better

(3) This paper models and measures the trust evidencefrom two aspects explicit feedback and implicit feed-back of customer to ensure maximum accuracy andintegrity We design direct trust indirect trust globaltrust and decision trust in IDTrust against individualmalicious and collusion node behaviors tomake trustmanagement more effective

In the remainder of the paper we introduce some relatedworks in the next section Section 3 describes the trust com-putation architecture (named IDTrust-Arch) which providesiterative and dynamic trust computation in P2P network forIDTrust Trust evidence modeling in IDTrust is proposedin Section 4 and trust measurement of IDTrust is listed inSection 5 The simulations and analysis follow in Section 6with conclusions afterwards in the last section

2 Related Work

Researchers have done lots of work around trust computingin decades Most of these achievements are about trust issuesin P2P network distributed ad hoc sensor network Internetand so forth Trust computing includes three modulesnamely trust evidence acquisition trust evaluation and trustinference We conclude current related work into four kindsin this section

(1) Trust Computing Based on Local Transaction EvidencesDirect trust or local trust expresses to what extent a trustorbelieves a trustee based on the trustorrsquos own local transactionhistory with the trustee Buchegger and Le Boudec proposedCONFIDANT protocol based on local information [9]and the model processes secondhand information directlythrough friends in P2P network or ad hoc network Thismethod could defend against single malicious behaviors butcould not avoid betrayal and collusion Damiani et al present

local trust model named XRep [10] they believe that nodesfrom same IP cluster are collusion group in P2P networkXRep has good convergence to recognize collusion behaviorsbut it has limitation and some honest nodes may also beincluded in the collusion IP cluster Jia et al [11] optimizedirect trust by power law and design forgetting factor basedon time Their model makes trust decision through a processof asking the neighborsrsquo feedback of the target node Thismodel is effective but not versatile because it assumes that thenodes must have trusted neighbors Direct trust has certaincognitive abilities on single malicious node but it has limita-tions on node collusion and malicious tampering evidence

(2) Trust Computing Based on Global Historical InformationGlobal trust is computed fromother nodes recommendationsto measure how trustworthy the target node is To computeglobal trust a trustor needs to aggregate all the trusteersquos trustevidences

There are two major ways to calculate global trust Thefirst way is based on central server calculation such as eBaywhich requires the peer nodes to deal with each other on thecentral platform The other way is fully distributed such asEigenTrust [12]which is proposed byKamvar et al EigenTrustreputation system can infer a unique global trust in a verydistributed way by history Such a global model does notneed an administration center and difficultly guarantees afast and secure convergencewhen computing the global trustNevertheless it inspires our works Dou et al [13] improvedthe EigenTrust system in computing convergence and modelsecurity However there remains an efficiency problem andits security mechanism is only from punishment and cer-tification Xiong and Liu [14] proposed a PeerTrust modelwith three basic trust parameters and two adaptive factorsand then defined a general trust metric to combine themefficiently Joslashsang et al [15] proposed a trust inferencemethodfor simplifying a complex network to express it in a series ofparallel networks This solution may lead to the loss of trustinformation They proposed an edge splitting method in thefurther works [16] to address this problem Nevertheless thismethod is valid only on a simple trust network and invalid oncomplex trust networks Wang and Nakao propose PoisonedWater [17] and Shaikh et al proposeGTMS [18] theymeasuretrust based on global group to improve the convergence rateof the trust calculation and global trust accuracy Song et alpropose global trust based on fuzzy logic inference rules [19]This method has higher detection rate of malicious nodesbut it can only avoid simple malicious behaviors and cannotdefend against various attacks on trust mechanism Gan et al[20] present a new method for trust recommendation Thismethod uses confidence factor to comprehend direct and rec-ommendation trust and establishes reward and punishmentmodel to encourage fair global recommendation

(3) Trust Computing Based on Global and Historical Informa-tion Correlation Correlation trust is based on global trustwith similarity or correlation factors Li et al [21] proposea global trust SWRTrust with cosine similarity which canidentify collaborative cheating malicious behaviors Das andIslam [22] propose an excellent dynamic trust computation

Mobile Information Systems 3

model SecuredTrust to cope with the strategically alteringbehavior of malicious agents and to distribute workload asevenly as possible among service providers Qiao et al [23]propose a novel network group behavioral model based ontrust by exploring behavior similarity aiming at perceptionand response of malicious network incidents The proposedmodel establishes trust relationship between nodes usinglarge scaled network topology and uses relevant trust conceptto increase trust value between weak correlations Othermodels like [24 25] also propose global trust with correlationcomputation to improve trust decision performance based onlarge scaled transaction histories

(4) Trust Computing Based on Inference Network with Multi-factors Another kind of trust computing model is based oninference network with probability or multidimensional fac-tors Kuter and Golbeck [26] propose a trust inference modelSUNNY based on trust network by evaluating trust and con-fidence with inference probability It has good effectivenessto discover trust recommender paths Vu and Aberer [27]use trust model to monitor dishonest behaviors via filteringunfair ratings in trust networkWang and Singh [28] proposetrust model based on evidences conflict probability And Canand Bhargava [29] propose a distributed trust computingalgorithm SORT where nodes create their own trust networkaccording to local trust information and infer the trust-worthiness of target nodes according to history interactionand recommendation information Liu et al [30] infer trustrelationships by small world theory and apply the proposedmodel in service network Tan et al [31] propose a novel trustinference model based on probabilistic model and balancetheory for P2P network and the proposed algorithm coulddiscover more valuable trust evidences paths for inferring thetargetrsquos trust They applied the proposed model in [32] andget a relatively effective inference performance Graduallyresearchers begin to infer trust degree withmultidimensionalevidence factors Wang and Wu [33] proposed a multidi-mensional evidence-based trust management system withmultitrusted paths (MeTrust for short) to conduct trust com-putation on any arbitrarily complex trusted graph The trustcomputation inMeTrust has three tiers namely the node tierthe path tier and the graph tier It is an excellent trust modelHowever it does not provide distributed storage structurefor P2P system Jiang and Li presented a novel reputation-based trust mechanism for P2P e-commerce systems [34]In this mechanism one peer has two kinds of reputationslocal reputations and global reputations To compute thelocal and global reputations precisely and to obtain strongerresistibility to attacks as well they use many comprehensivefactors in computing trust value in the mechanism Basicallythis model is a comprehensive mechanism However its timefactor is only linear and there is no clear method to resistteam malicious behaviors Tan et al [35] presented a globaltrust model with correlation factor based on communicationhistory and improved the time factor with exponentialequation It shows a rational history vector and presents threetrust models with multidimensional trust factors

Inspired by the above four kinds of trust computingmodel in this paper the proposed IDTrust has direct trust

indirect trust global trust and decision trust Meanwhilethe IDTrust aims to solve trust issues in large scaled P2Pwhich have dynamic and rapid evidence and to improve trustcomputing efficiency in iterative and dynamic computationbased on evidences with multifactors Some definitions andparameters have been discussed in our former work [24 3132 35] and IDTrust integrates methods and improves com-putation architecture

3 IDTrust-Arch

Trust measurement must be based on transaction or com-munication histories Large scaled P2P systems have highfrequency in generating data strong dynamics in transactionsand large scaled in transaction histories As a result trustevidences also have such same characteristics as high gener-ating speed and large scaled histories How to compute thesedynamic and large scaled trust evidences efficiently becomesa challenge to trust model

In IDTrust we firstly design a distributed iterative com-puting architecture named IDTrust-Arch as shown inFigure 1(a) including evidence collector (EC) and trust deci-sion (TD) which are designed extendedly based on P2P sys-tem architecture In IDTrust-Arch EC collects historic trans-actions from P2P system and feeds back evidence vectorto TD while TD computes the trustworthiness of objectivenodes according to evidence vector and feeds back trustdegree to P2P system according to request instructions Rela-tions of EC TD and original P2P service in IDTrust-Archhave the following features

(1) Independent and Extended Functions Within large scaledP2P system distributed trust computingmay causehuge com-munication traffic likely influencing the original P2P servicequality We mentioned this situation and it is under con-sideration in IDTrust-Arch Different from traditional P2Ptrust model which treats trust computation as an inner-binding service the proposed IDTrust-Arch is independentfrom P2P service Firstly EC TD and P2P service haveseparate communication ports and EC and TD do notoccupy P2P information channel For any peer node EC andTD are external extended functions but not internal bindingservice Secondly at any time EC TD and P2P service can bedeployed in different physicalmachines or different processesseparately

(2) Distributed but Topology Consistency As a result of theindependent functional design each node 119894 in P2P system canbe extended into three nodes as node(119894) EC(119894) and TD(119894) EC(and also TD) in all peer nodes can constitute a P2P networkindependently using same look-up routing algorithm as theoriginal P2P system Figure 1(d) shows the topology relationsbetween EC TD and the original P2P system As the figureshows the three will have the same topologies and commu-nicate with each other

(3) Iterative Computing IDTrust-Arch designs an iterativecomputing model for large scaled evidences in TD and ECwhich only computes latest evidence vector each time based


Evidence collector

Trust decision

Original P2P system

Historic transactions

Trust degree

Request

⟨Evidence⟩ Request

(a) Relations between EC TD and original P2P topology in IDTrust-Arch

Evidence⟨n⟩ UpdateTrust

evaluation

EC(i) TD(i) P2P(i)

fen

fenminus1gnminus1

gn

(b) Iterative computing model in IDTrust-Arch

Explicit evidence

User ratings⟨120593⟩

Implicit evidence

Succ rate⟨120582⟩

Decay⟨120588⟩

Similarity⟨sim⟩

EC(i)

Direct trust⟨dT⟩

Indirect trust⟨iT⟩

Global trust⟨gT⟩

Trans value⟨⟩

Decision trust⟨DeT⟩

TD(i)

P2P service

(c) Relations between evidences and trust model in IDTrust

P2P(i) TD(i) EC(i)

Node i

(d) Topology

Figure 1 IDTrust-Arch

on former result Figure 1(b) shows the iterative trust comput-ing processes We can see that each computation is using theresult of last computingThis kind of newmethod can greatlyimprove the trust computing efficiency and contains all theinformation of evidence by calculation of iterative accumu-lation comparing it to the traditional one which involvesall the evidence every time When large numbers of nodes

in P2P network are online dynamically for example some-times online sometimes offline traditional computingmodelresults in difference of trust evaluation due to the inconsis-tence of set of historic evidences (eg there are 100 evidencesof activity last time but may be only 30 next time) The itera-tive model in IDTrust-Arch can make use of evidences moreefficiently by keeping evaluation consistency all the timeThis


Table 1 Example cumulative evaluation value of 120593119899(1 2)

Iteration number 120593119905now (1 2) Time stamp 120572 120593

119899(1 2)

Initial Ratings from N1 to N2 1198791 = 10 1 01 095 1198792 = 20 1 0952 09 1198793 = 30 075 091253 07 1198794 = 80 0918367347 07173469394 08 1198795 = 100 0395061728 075

kind of trust computation of IDTrust-Arch can adapt to highspeed and large scaled P2P evidences flexibly The followingequation is the formalized formula of updating function

119891119890119899(119901 119902) = update (119891119890

119899minus1(119901 119902) 119892

119899(119901 119902))

119892119899(119901 119902) = update (119892

119899minus1(119901 119902) evidence (119901 119902))

(1)

Here we assume that node 119901 (called trustor) needs to calcu-late the trust degree of node 119902 (trustee) And119891119890

119899(119901 119902)denotes

the 119899th trust decision computing in node 119901 while 119892119899(119901 119902)

denotes the 119899th evidence modeling Appropriate initializeddata will be set for the very first iterative computing The fol-lowing IDTrustmodeling is based on IDTrust-Arch and (1) Ithas been noted that (1) is only a formalized iterative computa-tion method to guide modeling in IDTrust and actual modelsymbols and parameters depend on modelrsquos requisitions

4 Trust Evidence Modeling for IDTrust

Trust evidences in IDTrust come from EC including explicitfeedback evidence and implicit feedback evidenceThe expli-cit feedback evidence is from customer nodes active evalu-ation while implicit feedback evidence should be excavatedfrom transactions We formalize trust evidence vector inIDTrust as Evidence(119901 119902) to represent evidences that node119901 has upon node 119902 That is modeled by

Evidence (119901 119902) = ⟨⟨120593⟩ ⟨120582 ] 120588 sim⟩⟩ (2)

Explicit evidence vector in IDTrust is mainly consisting ofuser ratings 120593 and of course it could be further expandedto multidimensional evidence vector in specific P2P sys-tems Implicit evidence vector in IDTrust mainly consistsof transaction success rate 120582 transaction value ] (based ontransaction price) trust decay factor 120588 and nodes similaritysim The following will model both explicit evidence vector⟨120593⟩ and implicit evidence vector ⟨120582 ] 120588 sim⟩ in iterative styleaccording to (1)

41 Explicit Evidence Usually explicit evidence is fed back byuser ratings or satisfactions In practice the quantitative orqualitative feedback can be by scores (5 points or 100 points)or by satisfaction (satisfied mostly satisfied not satisfiedetc) In this paper we use user evaluation by range [0 1] tounify usersrsquo ratings or satisfactions

Definition 1 User evaluation is denoted by 120593119905

(119901 119902) isin [0 1]

and represents that user 119901 evaluated user 119902with a [0 1] rating

t1

1 minus 120572 120572

tnminus1 tn

S(Δ) = 1

Δt = (tnminus1 minus t1)Δt = (tn minus t1)

Figure 2 Dynamic factor of current evaluation

after transaction with 119902 at the time 119905 120593119905(119901 119902) = 0 representsthat node 119901 is fully unsatisfied with node 119902rsquos service while120593119905

(119901 119902) = 1 represents being completely satisfied Let120593119905now (119901 119902) mean the latest user evaluation at the time 119905now

while 119905 = 1199051means the very beginning evaluation time

Definition 2 Using 120593119899(119901 119902) to denote the current cumulative

evaluation value by node 119901 upon node 119902 based on the eval-uation of last time according to iterative model assume1205930(119901 119902) = 0 that is

120593119899(119901 119902) = 120572 sdot 120593

119905now (119901 119902) + (1 minus 120572) sdot 120593119899minus1

(119901 119902) (3)

where theweight120572 is the dynamic factor of current evaluation120593119905now (119901 119902) In IDTrust 120572 will be dynamically changed by the

distance between 119905119899minus1

and 119905119899 and larger distance will bring

bigger 120572 as shown in Figure 2 In other words the currentevaluation 120593119905now (119901 119902)will be more important if user 119901 did nottransact with 119902 after a longer time

In Figure 2 we calculate 120572 by the method of right trianglewhere area is 1 (unit area) and the base is global timedifference (from 119905

1to 119905now) of interaction between 119901 and 119902

Therefore assume 120572 = 1 at the very beginning (119905 = 1199051) the

weight 120572 of current evaluation is the area of the triangle fromtime 119905

119899minus1to 119905119899 That is

120572 =

1 minus (119905119899minus1

minus 1199051

119905119899minus 1199051

)

2

if 119905119899gt 1199051

1 if 119905119899= 1199051

(4)

For example assume there are four transactions that hap-pened between node 1198731 and node 1198732 just as shown inTable 1 Then the cumulative evaluation value would be com-puted according to (3) based on evaluations from1198731 to1198732while 120572 changed based on transaction time stamps After fouriterations the value of 120593

119899(1 2) is 075 This value is less than

the average of transaction rations from1198731 to1198732 because ofthe dynamic changes of 120572


42 Implicit Evidences

421 Successful Rate Successful transaction is a positivestimulus for nodes while failure of transaction (or other sit-uations such as complaint) would result in a negative growthto nodes trust

Definition 3 120582119899(119901 119902) denotes the percentage of cumulative

successful transaction of node 119901 to node 119902 Assume 1205820(119901 119902) =

0 That is

120582119899(119901 119902) =

(119899 minus 1) 120582119899minus1

(119901 119902) + 120582119905now (119901 119902)

119899 (5)

where 120582119905now (119901 119902) = 0 if successful 1 if failed represents

whether the current 119899th transaction is successful or not

422 Transaction Value Transaction amount (in real orvirtual currency) is typical implicit trust evidence Peopleusually think that the service provider with larger transactionamount is more valuable to trust in intuition However peo-ple are also worried about the malicious or dishonest blocktrade which may be only baits for honest users Thus wemodel the transaction amount as transaction value which isbased on both transaction amount and successful rate

Definition 4 Use ]119905now (119901 119902) = 120596119905now (119901 119902) sdot 120582

119899(119901 119902) to denote

the current transaction value of the current user transactionamount 120596119905now (119901 119902)

For instance if the current transaction amount120596119905now (119901 119902)is 100 but the current successful rate 120582

119899(119901 119902) is only 025 then

we take the current transaction value to be only 100 lowast 025 =

25 This engagement can stimulate honest users to improvetheir transaction successful rate 120582 and also suppress the mali-cious behaviors which strategically raise the trustworthinessby transaction amounts meanwhile

Definition 5 Using ]119899(119901 119902) = ]119905now (119901 119902) + ]

119899minus1(119901 119902)

to denote cumulative transaction value of the transactionamount assume ]

0(119901 119902) = 0

423 Time Decay Factor for Trustworthiness In the processof trust evaluation the more recent the trust evidence is themore important it is If a user node takes a long time withouttransacting with other nodes its trust will be degradedgradually We apply exponential function to design the decayfactor of direct trustworthiness of nodes which will be usedin direct trust in IDTrust That is

120588 = 119890minus120581Δ119905

(6)

Here 119896 (119896 gt 0) is the decay speed and Δ119905 = 119905119899minus 119905119899minus1

denotes the time difference for a certain user node Figure 3 isthe illustration of decay curves of four different decay speedsApparently the lager 119896 is the faster the decay would be

424 Similarity of User Evaluations Similarity is one kindof important implicit evidence which indicates the extent towhich two nodes are alike especially when there is no direct

Time decay curve

k = 050

k = 010

k = 005

k = 001

100 200 300 400 5000Δt

0

01

02

03

04

05

06

076

08

09

1

Tim

e dec

ay v

alue

Figure 3 Time decay curves

transaction between two strange nodes In order to measurethe similarity between user nodes 119901 and 119902 we choose thethird set of nodes that have interaction with both of them tobe a computing object For example node 119860 did not know119861 but both share common friends 119862119863 and then 119860 canjudge similar correlation with 119861 via its friends 119862119863 If thecommunication history among 119860 sim 119862119863 is similar to thehistory among 119861 sim 119862119863 then we think 119860 and 119861 have verysimilar correlation

Let 119868119909denote the set of nodes which node 119909 has ever

interacted with Let 119868119901119902

= 119868119901cap 119868119902denote the common

third-part set which interacts with both users 119901 and 119902 Thesimilarity computation is based on user-user matrix 119877(119899 times 119899)

whose element 119903119894119895is

119903119894119895=

120593119899(119894 119895) if 119894 = 119895

1 if 119894 = 119895

(7)

If no user interaction happened between user nodes 119894 and119895 we assume 119903

119894119895= 0 Then

119877 = (119903119894119895)119899times119899

=

[[[[[

[

1 11990312

sdot sdot sdot 1199031119899

11990321

1 sdot sdot sdot 1199032119899

sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot sdot

1199031198991

1199031198992

sdot sdot sdot 1

]]]]]

]

(8)

Each row 119894 in 119877 denotes evaluation to all other user nodesfrom node 119894 There are many methods to calculate the simi-larity between two items such as cosine similarity correlationsimilarity and adjusted cosine similarity Cosine similarityis a general method to compute the similarity between twouser nodes However pure cosine formula cannot distinguishthe subjective difference of evaluation criteria that differentusers have for the same object For instance for a certain user119909 user 119901 evaluated 119909 with value 09 after a transaction anduser 119902 evaluated it with value 06 in another transaction Theevaluated value is quite different however it is probable that


09 and 06 both stand for ldquosatisfiedrdquo separately in perspectiveof 119901 and 119902 and the difference may only be because user 119902has more strict evaluation criteria In this situation ordinarycosine formula is not appropriateThus we apply the adjustedcosine formula that is subtracted by average value to computethe similarity based on matrix 119877 That is

119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2

sdot radicsum119888isin119868119901119902

(120593119899(119902 119888) minus 120593

119899(119888))2

(9)

where 120593119899(119888) denotes the average evaluation value of node 119888 isin

119868119901119902In the above similarity equation three questions need to

be further considered

(1) Sparse Common Nodes It is a challenge when |119868119901119902| is zero

or very smallThis phenomenonmay even result in similaritybeyond computing Thus in IDTrust we assume that simi-larity should be computed by cosine only when |119868

119901119902| reaches

threshold 120591 (natural number that is larger than 1)

(2) Result Adjusting Although the range of user evaluation 120593

is within [0 1] in IDTrust due to inner product of deviationthe return value of adjusted cosine similarity by (9) is within[minus1 1] where range [minus1 0] denotes the evaluation vectorsangle of 119901 and 119902 is between degrees [90 180] which meansthe two vectors are completely irrelevant (orthogonal) or evenoppositeThus we assume it is completely dissimilarity in thiscase and the similarity is set to zero when 119888 sim

119899(119901 119902) return

[minus1 0] in IDTrust As a result the range of similarity will beadjusted to [0 1] which is easier to control

(3) Default Similarity At the very beginning (or when sim-ilarity is beyond computing) we need to initialize a defaultsimilarity In IDTrust the default similarity is 05

Hence (9) is modified to be

sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)

Alsowe assume the similarity of sim119899(119909 119909) = 1 whichmeans

the similarity among nodes is reflexive Note that similarityamong nodes is symmetrical that is sim

119899(119901 119902) = sim

119899(119902 119901)

Therefore similarity among nodes denotes the compatibilityrelationship It indicates that we can get a similar set in thewhole set by compatibility relationship which may be a newway to get complete coverage for similar nodes in the wholeset in our further research But in IDTrust now we have notapplied this feature

p q

dT(p q)

(a) Trust evaluation directlyc

qpiT(p q)

dT(c p)dT(p c)rarr sim(p c)

(b) Trust evaluation indirectly

Figure 4 Directindirect trust computing

5 Iterative Trust MeasurementModels in IDTrust

Based on the above iterative computing architecture and evi-dence modeling we design direct trust model indirect trustmodel global trust model and decision trust model forIDTrust against malicious behaviors especially collusionbehaviors Figure 1(c) shows relations between trust evi-dences and trust measurement models

51 Direct Trust Direct trust (also called local trust) is basedon the direct transaction experiences between the trustorand trustee as Figure 4(a) shows We define the direct trust-worthiness of node 119901 to node 119902 from user evaluation whichis from the direct transaction between the two nodes LetdT119899(119901 119902) isin [0 1] denote the direct trust of node 119901 to node 119902

That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)

As you can see the better the service that node 119902providedthe higher the direct trust degree that node 119902 obtains fromnode 119901

52 Indirect Trustworthiness To compute indirect trust bytraditional trust models the trustor needs to aggregate all ofthe trusteersquos transaction evaluations from recommenders andthen compute or infer indirect trust degree of target node bythose recommendersrsquo evidences Figure 4(b) illustrates thatnode 119901 calculates indirect trust degree of node 119902 throughrecommender node 119888

As we know we can obtain the direct trust dT(119888 119902) fromrecommender node 119888 and direct trust dT(119901 119888) from 119901 Somerelated indirect trust models use these two kinds of directtrust to calculate indirect trust of the target node such asdT(119901 119888) lowast dT(119888 119902) or other similar forms However sincethe direct trust is usually based on user evaluation and theuser evaluation is subjective in a certain extent different usershave different evaluation criteria just as we discussed inSection 424 about similarity of user evaluations aboveThuswe consider user similarity between nodes 119901 and 119888 insteadof direct trust of 119901 and 119888 for the similarity computation inIDTrust is also based on user evaluation (which is also equiv-alent to direct trust) Using similarity can also avoid collusion


Inputresource key words

P2P lookup inservice providers

Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet

Returnthe first q in decTList

No

Yes

Node q

Servicerequest

Communicate withservice provider

decTListadd(decT(p q))Put decT(p q) into decTlist

decT(p q) = globalTrust(p q q)) lowast (p

Is q the last one

Figure 5 Trust decision process

malicious recommenders since malicious evaluation wouldquite differ from the honest evaluation

Let iT119899(119901 119902) isin [0 1] denote indirect trust of node 119901 to

node 119902 which is calculated by direct trust of recommenderset 119888Set to node 119902 and use the similarity between node 119901 and119888Set as another indirect parameter That isiT119899(119901 119902)

=

radicsum119888isin119888Set sim119899 (119901 119888) sdot dT119899 (119888 119902)

|119888Set| if |119888Set| gt 0

0 otherwise

(12)

where if recommender nodes set 119888Set is empty the indirecttrust return is zero In order to read the result clearly theabove formula amplifies the valuewith square root since bothsimilarity and direct trust ranged within [0 1]

53 Global Trust Now we define the global trust as theweighted sum of direct trust and indirect trust where thegeneral weight is the decay factor 120588

gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))

+ (1 minus 120588) sdot gT119899minus1

(119901 119902)

(13)

Here 120577 isin [0 1] is the weight of the direct trust withdefault value 05 which could be adjusted by users accordingto the requirement

54 Decision Trust Decision trust is an integrated value builtfrom iterative global trust and implicit evidence of transac-tion value to represent a comprehensive expression for usernodesrsquo trust

Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]

119899(119901 119902) denote decision trust

which is based on global trust and implicit evidence transac-tion value For example if gT

1010(1 3) = 065 and transaction

value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65

indicating the integrated trust of node 3 is 65 from theperspective of node 1 As we can see the gT

119899(119901 119902) is a global

factor while ]119899(119901 119902) is a relatively local factor we apply these

two factors to help in decision makingHereby we also design a trust decision algorithm as

shown in Figure 5 In the algorithm process as a trustor usernode 119901 initiates the trust decision in order to find the mosttrustworthy nodes At the beginning node 119901 inputs resourcekeywords to request P2P services and then P2P networkreturns a service provider list (119902Set) specifying who canprovide related resource services Each decision trust (decT)will be calculated through EC and TD and all the decisiontrust will be ranked in descending order At last the node 119902that has the highest decision trust valuewill be recommendedto node 119901

6 Simulations and Analysis

In order to prove the rightness and efficiency of IDTrust wedesign a simulation system based on IDTrust-Arch and ourformer work We will separately verify IDTrust performanceagainst dynamic individual malicious and collusion mali-cious environment and verify the iterative computation per-formance comparing with traditional model

61 Simulation Environment Firstly we design a P2P com-munication architecture as in Figure 6 based on IDTrust-Arch introduced above In the system service provider (SP)


Extended interfaceIF⟨EC SP⟩

Service provider (SP)

Service requester (SR)

Original P2P interfaceIF⟨P2P SP SR⟩

Evidence collector (EC)

Trust decision (TD)

Request

Response

Extended interfaceIF⟨TD SP⟩

Request

Requ

est

Response

Resp

onse

Inner interfaceIF⟨TD EC⟩

Request

Response

Node y

Extension part

Node x

Figure 6 Communication architecture of simulation system

is the node that provides services while service requester (SR)needs the resource services EC is to collect and integrate thehistoric data and obtain evidence vector from P2P systemTD is to compute the trustworthiness of objective nodesaccording to evidence vector provided by EC The origi-nal P2P interface IF⟨P2P SP SR⟩ manages the communica-tion between node 119909 and node 119910 The extended interfaceIF⟨TD SP⟩ manages the communication between SP andTD And interface IF⟨EC SP⟩ manages the communicationbetween SP and ECThe inner interface IF⟨TDEC⟩managesthe communication between TD and EC

We implement the above simulation system based onMSNET Framework 45 and MS P2P Networking platformwith 119888 programming language thread pool technique anddistributed communication In topology we use simplifiedChord to be the P2P routing algorithm

We simulate three kinds of P2P nodes similar to [24 35]

(1) Node of Class A It is the normal node in P2P systemwhichprovides normal service and evaluates service almost equal tosimulated resource value

(2) Node of Class B It is an individual malicious node thatprovides false service andmakes fault evaluation for any othernodes at random independently

(3) Node of Class C It belongs to some malicious team nodesproviding false service and evaluates normal nodes with faultfeedback It should be noted that it overstates teammembersrsquoservices with high score to cheat other nodes out of theirteam

We simulated 1000 nodes and each node has 50sim100resources abstract and designed 4 kinds of experiment andalmost 1000000 simulated transactions happened betweennodes Table 2 shows the nodes proportion allocation fordifferent experiments

Table 3 initializes different parametersrsquo value Most ofthese parameters will be dynamically changed with the itera-tive computation going on

Table 2 Nodes proportion setup in different experiments

Experiment Parameter AllocationExperiment 1IDTrust against individual maliciousbehaviors

A nodeB nodeC node

75250

Experiment 2IDTrust against collusion maliciousbehaviors

A nodeB nodeC node

75025

Experiment 3Iterative performanceExperiment 4Trust decision performance

A nodeB nodeC node

701515

Table 3 Initialized parameters

Parameter Initializedvalue Description

119873 1000 The number of simulated nodes1205930 0 Initialized value of user evaluation

1205820 0 Initialized value of successful transaction

rate]0 0 Initialized value of transaction valuesim0 05 Initialized value of similarity

iT0 0 Initialized value of indirect trust value

gT0 05 Initialized value of global trust value

120572 1 Initialized value of userrsquos currentevaluation

119896 001 Decay speed120591 10 Threshold of |119868

119901119902|

120577 05Weight of direct trust which could beadjusted by users according torequirement

62 Simulation and Analysis

Experiment 1 (IDTrust against individual malicious behav-iors) The first adversary model for IDTrust is individualmalicious nodeWe set up almost 25of individualmalicious


Global trust trend for nodes from Class A and Class B

a1 honest node from Class Ab1 individual malicious node from Class B

20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree

(a) IDTrust against individual malicious behaviors

Global trust trend for nodes from Class A and Class C

a2 honest node from Class Ac1 collusion malicious node from Class C

20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr

ee(b) IDTrust against collusion malicious behaviors

Figure 7 Simulation for defending against malicious attacks

Noniterative trust computation time

50 100 1500 200Trust evidences (times100)

0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)

(a) Noniterative trust computation time

Iterative trust computation timetimes10minus3

20 40 60 80 100 120 140 160 180 2000Trust evidences (times100)

0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)

(b) Iterative trust computation time

Figure 8 Iterative computation performance comparing with noniterative computation

nodes of Class B and 75 of normal nodes From observationof IDTrust global trust value of designed honest node ofClassA and individualmalicious node of Class B during simulationexperiment we can get result as shown in Figure 7(a) Withincreasing of iterative time global trust value of observedindividual malicious node decays very fast in contrast trustvalue of honest node increases all the way This means thatdefense to individual malicious node is effective

Experiment 2 (IDTrust against collusion malicious behav-iors) In this experiment we set up 25 of collusion mali-cious nodes of Class C and 75 of normal nodes We want tosee whether IDTrust can defend against collusion mali-cious behaviors Figure 7(b) shows results At the beginning

IDTrust cannot distinguish and recognize collusion mali-cious node due to historic set of public transactions Withiterative calculation increasing the trust value of collusionnode degraded drastically to almost zero It indicates the sim-ilarity in IDTrust workswell and IDTrust is effective to defendcollusion behaviors

Experiment 3 (performance of iterative calculation) In orderto verify the efficiency of iterative calculation in IDTrust wecount the computation time cost of decision trust computa-tion statistically in both iterative andnoniterative calculationsseparately In this experiment we set up 15 of collusionmalicious nodes 15 of individual malicious nodes and 70of normal nodes Figure 8(a) shows the result of noniterative


0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1

)Performance of resisting transaction with malicious nodes

IDTrustMDHTrustNBRTrust

EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations

Figure 9 Trust decision performance comparisons

calculation time cost while Figure 8(b) is about iterativecalculation in IDTrust As we can see from the comparisonthe IDTrust with iterative calculation improves computationperformance greatly and its time cost is much smallerthan that of noniterative one The experiment indicates thatarchitecture of IDTrust is highly efficient

Nevertheless this iterative performance still occupiedCPU and memory utilization During the simulation theCPU (Intel Core i7-4980HQ28GHz) utilization is almost60 and the memory (4G DDR3 RAM) utilization is 73or so But this disadvantage would be helpful for our furtherstudy

Experiment 4 (performance of trust decision) Finally in thispaper we compare performance of trust decision of IDTrustwith EigenTrust [12] MDHTrust [35] NBRTrust [24] andrandommodel with same nodes proportion as Experiment 3In this experiment we count the number of malicious nodesof Class B and Class C which are selected to be trustees bynormal nodes of Class A Also we define rejectRate to denoteto what extent the normal nodes reject services against mali-cious nodes

rejectRate = Count of transacted malicious nodesCount of all transacted nodes

(14)

Experiment result is shown in Figure 9 As we can seeIDTrust has rapid convergence to reject transactions withmalicious nodes since it has strict trust decision processesas described in Section 54 The random trust model hasrelatively lowest performance for it has no trust decision def-inition This experiment indicates that IDTrust is very sensi-tive to malicious node and has an advantage of recognizingmalicious behavior

7 Conclusions

In this paper we propose a novel dynamic trust model withiterative computation for P2P systems according to dynamicsand fast responding characteristics in large scaled P2P net-work named IDTrust

Firstly we propose a distributed computing architecturenamed IDTrust-Arch to obtain trust evidences and maketrust decision including three computing modules ECTC and original P2P module which are independent butcommunicate with each other to get higher computationperformanceThis architecture separates the trust computingtask from P2P system while traditional trust computing is abinding service and based on universal set of evidence factsto obtain global trust degree of a given node

Secondly we propose an iterative computation methodto model trust evidence vector and calculate trust inIDTrustThis design degrades the trust computation cost andimproves the computation performance Based on the aboveIDTrust is proposed Trust evidence vector in IDTrust inclu-des both explicit and implicit trust evidences to improve theevidence integrity Direct trust indirect trust global trustand decision trust are designed in IDTrust based on explicittransaction evaluation and implicit evidence like transactionvalue successful rate decay factor and similarity factor

Finally we design a simulation system based on IDTrust-Arch Simulations against both individual and collusionmalicious nodes prove that the IDTrust is right and efficientagainst the two adversary models Simulations about iterativecomputation and trust decision prove that IDTrust has highcomputation performance and more efficient trust decisionperformance

Conflict of Interests

The authors of this paper declare that they have no conflict ofinterests

Acknowledgments

This work is supported by the National Natural ScienceFoundation of China under Grant no 61402097 and no61572123 the National Science Foundation for DistinguishedYoung Scholars of China under Grant no 61225012 and no71325002 the Specialized Research Fund of the DoctoralProgram of Higher Education for the Priority DevelopmentAreas under Grant no 20120042130003 The authors alsoacknowledge Ayebazibwe Lorrettarsquos help in doing the gram-matical revision

References

[1] H Esaki ldquoA consideration on RampD direction for future Internetarchitecturerdquo International Journal of Communication Systemsvol 23 no 6-7 pp 694ndash707 2010

[2] J Duarte S Siegel and L Young ldquoTrust and credit the roleof appearance in peer-to-peer lendingrdquo Review of FinancialStudies vol 25 no 8 pp 2455ndash2484 2012

[3] R Emekter Y Tu B Jirasakuldech and M Lu ldquoEvaluatingcredit risk and loan performance in online Peer-to-Peer (P2P)lendingrdquo Applied Economics vol 47 no 1 pp 54ndash70 2014

[4] E Lee and B Lee ldquoHerding behavior in online P2P lendingan empirical investigationrdquo Electronic Commerce Research andApplications vol 11 no 5 pp 495ndash503 2012


[5] Z Jie ldquoA survey on trust management for VANETsrdquo in Proceed-ings of the 25th International Conference on Advanced Informa-tionNetworking andApplications pp 105ndash112 SingaporeMarch2011

[6] L Mekouar Y Iraqi and R Boutaba ldquoPeer-to-peerrsquos mostwanted malicious peersrdquo Computer Networks vol 50 no 4 pp545ndash562 2006

[7] H Q Lin Z T Li and Q F Huang ldquoMultifactor hierarchicalfuzzy trust evaluation on peer-to-peer networksrdquo Peer-to-PeerNetworking and Applications vol 4 no 4 pp 376ndash390 2011

[8] C Selvaraj and S Anand ldquoA survey on security issues ofreputation management systems for peer-to-peer networksrdquoComputer Science Review vol 6 no 4 pp 145ndash160 2012

[9] S Buchegger and J Y Le Boudec ldquoPerformance analysis of theCONFIDANTprotocolrdquo in Proceedings of the 3rd ACM Interna-tional Symposium on Mobile Ad Hoc Networking amp Computing(MobiHoc rsquo02) pp 226ndash236 New York NY USA June 2002

[10] E Damiani S D C Vimercati and S Paraboschi ldquoA reputa-tion-based approach for choosing reliable resources in peer-to-peer networksrdquo in Proceedings of the 9th ACM Conferenceon Computer and Communications Security pp 207ndash216 ACMPress Washington DC USA November 2002

[11] C Jia L Xie X GanW Liu and ZHan ldquoA trust and reputationmodel considering overall peer consulting distributionrdquo IEEETransactions on Systems Man and Cybernetics Part A Systemsand Humans vol 42 no 1 pp 164ndash177 2012

[12] S D Kamvar M T Schlosser and H Garcia-Molina ldquoTheEigentrust algorithm for reputation management in P2P net-worksrdquo in Proceedings of the 12th International Conference onWorld Wide Web (WWW rsquo03) pp 640ndash651 ACM May 2003

[13] W Dou H M Wang and Y Jia ldquoA recommendation-basedpeer-2-peer trust modelrdquo Journal of Software vol 15 no 4 pp571ndash583 2004

[14] L Xiong and L Liu ldquoPeerTrust supporting reputation-basedtrust for peer-to-peer electronic communitiesrdquo IEEE Transac-tions on Knowledge and Data Engineering vol 16 no 7 pp 843ndash857 2004

[15] A Joslashsang R Hayward and S Pope ldquoTrust network analysiswith subjective logicrdquo in Proceedings of the 29th AustralasianComputer Science Conference (ACSC rsquo06) vol 48 pp 85ndash94Hobart Australia January 2006

[16] A Joslashsang and T Bhuiyan ldquoOptimal trust network analysiswith subjective logicrdquo in Proceedings of the 2nd InternationalConference on Emerging Security Information Systems and Tech-nologies (SECURWARE rsquo08) pp 179ndash184 Cap Esterel FranceAugust 2008

[17] Y Wang and A Nakao ldquoPoisonedwater an improved approachfor accurate reputation ranking in P2P networksrdquo FutureGeneration Computer Systems vol 26 no 8 pp 1317ndash1326 2010

[18] R A Shaikh H Jameel B J drsquoAuriol H Lee S Lee and Y-J Song ldquoGroup-based trust management scheme for clusteredwireless sensor networksrdquo IEEE Transactions on Parallel andDistributed Systems vol 20 no 11 pp 1698ndash1712 2009

[19] S Song K Hwang R F Zhou and Y-K Kwok ldquoTrusted P2Ptransactions with fuzzy reputation aggregationrdquo IEEE InternetComputing vol 9 no 6 pp 24ndash34 2005

[20] Z-B Gan Q Ding K Li and G-Q Xiao ldquoReputation-basedmulti-dimensional trust algorithmrdquo Journal of Software vol 22no 10 pp 2401ndash2411 2011

[21] J-T Li Y-N Jing X-C Xiao X-P Wang and G-D Zhang ldquoAtrust model based on similarity-weighted recommendation for

P2P environmentsrdquo Journal of Software vol 18 no 1 pp 157ndash167 2007

[22] A Das and M M Islam ldquoSecuredTrust a dynamic trustcomputation model for secured communication in multiagentsystemsrdquo IEEE Transactions on Dependable and Secure Comput-ing vol 9 no 2 pp 261ndash274 2012

[23] L Qiao H Hui F Bingxing Z Hongli and W YashanldquoAwareness of the network group anomalous behaviors basedon network trustrdquo Chinese Journal of Computers vol 37 no 1pp 1ndash14 2014

[24] Z Tan H Wang W Cheng and G Chang ldquoA distributedtrust model for P2P overlay networks based on correlativityof communication historyrdquo Journal of Northeastern University(Natural Science) vol 30 no 9 pp 1245ndash1248 2009

[25] J Yin Z-S Wang Q Li and W-J Su ldquoPersonalized recom-mendation based on large-scale implicit feedbackrdquo Journal ofSoftware vol 25 no 9 pp 1953ndash1966 2014

[26] U Kuter and J Golbeck ldquoUsing probabilistic confidencemodelsfor trust inference in web-based social networksrdquo ACM Trans-actions on Internet Technology vol 10 no 2 article 8 23 pages2010

[27] L Vu and K Aberer ldquoEffective usage of computational trustmodels in rational environmentsrdquo ACM Transactions onAutonomous and Adaptive Systems vol 6 no 4 article 24 25pages 2011

[28] YWang andM P Singh ldquoEvidence-based trust amathematicalmodel geared for multi agent systemsrdquo ACM Transactions onAutonomous and Adaptive Systems vol 5 no 4 article 14 28pages 2010

[29] A B Can andB Bhargava ldquoSORT a self-organizing trustmodelfor peer-to-peer systemsrdquo IEEETransactions onDependable andSecure Computing vol 10 no 1 pp 14ndash27 2013

[30] F M Liu L Wang L Gao H X Li H F Zhao and S K MenldquoA Web Service trust evaluation model based on small-worldnetworksrdquo Knowledge-Based Systems vol 57 pp 161ndash167 2014

[31] Z Tan L Zhang and G Yang ldquoBPTrust a novel trust inferenceprobabilistic model based on balance theory for peer-to-peernetworksrdquo Elektronika ir Elektrotechnika vol 18 no 10 pp 77ndash80 2012

[32] Z H Tan G M Yang and W Cheng ldquoDistributed trustinference model based on probability and balance theory forpeer-to-peer systemsrdquo International Journal on Smart Sensingand Intelligent Systems vol 5 no 4 pp 1063ndash1080 2012

[33] G Wang and J Wu ldquoMulti-dimensional evidence-based trustmanagement with multi-trusted pathsrdquo Future GenerationComputer Systems vol 27 no 5 pp 529ndash538 2011

[34] S-X Jiang and J-Z Li ldquoA reputation-based trust mechanismfor P2P e-commerce systemsrdquo Journal of Software vol 18 no10 pp 2551ndash2563 2007

[35] Z-H Tan X-WWangW Cheng G-R Chang and Z-L ZhuldquoA distributed trust model for peer-to-peer networks based onmulti-dimension-history vectorrdquo Chinese Journal of Computersvol 33 no 9 pp 1725ndash1735 2010

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



large scaled P2P systems First of all explicit and implicitevidences are defined in IDTrust and then direct trustindirect trust global trust and decision trust are modeled inmathematical expressions based on evidencesThis paper hasthe following innovative features

(1) How to obtain large number of trust evidences andmake metric decision iteratively and dynamically isthe key in trust computing within an environmentof large scale P2P Internet As a result we proposea kind of computing architecture with hierarchicaldistribution trust which layers these three parts inde-pendently evidence collection trust computing andP2P communication leading to a three-layered P2Ptopology and enhancing the distribution computingof P2P commercial communication

(2) Traditionally each trust computing involves all thetrust evidence which contradicts the characteristicof P2P rapid and large scaled data Therefore in thispaper we propose an iterative and dynamic trustcomputing model Within this model trust comput-ing is iterative and only requires the newest trust evi-dence The performance of trust computing is better

(3) This paper models and measures the trust evidencefrom two aspects explicit feedback and implicit feed-back of customer to ensure maximum accuracy andintegrity We design direct trust indirect trust globaltrust and decision trust in IDTrust against individualmalicious and collusion node behaviors tomake trustmanagement more effective

In the remainder of the paper we introduce some relatedworks in the next section Section 3 describes the trust com-putation architecture (named IDTrust-Arch) which providesiterative and dynamic trust computation in P2P network forIDTrust Trust evidence modeling in IDTrust is proposedin Section 4 and trust measurement of IDTrust is listed inSection 5 The simulations and analysis follow in Section 6with conclusions afterwards in the last section

2 Related Work

Researchers have done lots of work around trust computingin decades Most of these achievements are about trust issuesin P2P network distributed ad hoc sensor network Internetand so forth Trust computing includes three modulesnamely trust evidence acquisition trust evaluation and trustinference We conclude current related work into four kindsin this section

(1) Trust Computing Based on Local Transaction EvidencesDirect trust or local trust expresses to what extent a trustorbelieves a trustee based on the trustorrsquos own local transactionhistory with the trustee Buchegger and Le Boudec proposedCONFIDANT protocol based on local information [9]and the model processes secondhand information directlythrough friends in P2P network or ad hoc network Thismethod could defend against single malicious behaviors butcould not avoid betrayal and collusion Damiani et al present

local trust model named XRep [10] they believe that nodesfrom same IP cluster are collusion group in P2P networkXRep has good convergence to recognize collusion behaviorsbut it has limitation and some honest nodes may also beincluded in the collusion IP cluster Jia et al [11] optimizedirect trust by power law and design forgetting factor basedon time Their model makes trust decision through a processof asking the neighborsrsquo feedback of the target node Thismodel is effective but not versatile because it assumes that thenodes must have trusted neighbors Direct trust has certaincognitive abilities on single malicious node but it has limita-tions on node collusion and malicious tampering evidence

(2) Trust Computing Based on Global Historical InformationGlobal trust is computed fromother nodes recommendationsto measure how trustworthy the target node is To computeglobal trust a trustor needs to aggregate all the trusteersquos trustevidences

There are two major ways to calculate global trust Thefirst way is based on central server calculation such as eBaywhich requires the peer nodes to deal with each other on thecentral platform The other way is fully distributed such asEigenTrust [12]which is proposed byKamvar et al EigenTrustreputation system can infer a unique global trust in a verydistributed way by history Such a global model does notneed an administration center and difficultly guarantees afast and secure convergencewhen computing the global trustNevertheless it inspires our works Dou et al [13] improvedthe EigenTrust system in computing convergence and modelsecurity However there remains an efficiency problem andits security mechanism is only from punishment and cer-tification Xiong and Liu [14] proposed a PeerTrust modelwith three basic trust parameters and two adaptive factorsand then defined a general trust metric to combine themefficiently Joslashsang et al [15] proposed a trust inferencemethodfor simplifying a complex network to express it in a series ofparallel networks This solution may lead to the loss of trustinformation They proposed an edge splitting method in thefurther works [16] to address this problem Nevertheless thismethod is valid only on a simple trust network and invalid oncomplex trust networks Wang and Nakao propose PoisonedWater [17] and Shaikh et al proposeGTMS [18] theymeasuretrust based on global group to improve the convergence rateof the trust calculation and global trust accuracy Song et alpropose global trust based on fuzzy logic inference rules [19]This method has higher detection rate of malicious nodesbut it can only avoid simple malicious behaviors and cannotdefend against various attacks on trust mechanism Gan et al[20] present a new method for trust recommendation Thismethod uses confidence factor to comprehend direct and rec-ommendation trust and establishes reward and punishmentmodel to encourage fair global recommendation

(3) Trust Computing Based on Global and Historical Informa-tion Correlation Correlation trust is based on global trustwith similarity or correlation factors Li et al [21] proposea global trust SWRTrust with cosine similarity which canidentify collaborative cheating malicious behaviors Das andIslam [22] propose an excellent dynamic trust computation






3 IDTrust-Arch







Evidence collector

Trust decision

Original P2P system


Trust degree

Request




evaluation

EC(i) TD(i) P2P(i)

fen

fenminus1gnminus1

gn


Explicit evidence


Implicit evidence


Decay⟨120588⟩

Similarity⟨sim⟩

EC(i)




Trans value⟨⟩


TD(i)

P2P service


P2P(i) TD(i) EC(i)

Node i

(d) Topology







119899(1 2)



119891119890119899(119901 119902) = update (119891119890

119899minus1(119901 119902) 119892

119899(119901 119902))

119892119899(119901 119902) = update (119892

119899minus1(119901 119902) evidence (119901 119902))

(1)


119899(119901 119902)denotes





Evidence (119901 119902) = ⟨⟨120593⟩ ⟨120582 ] 120588 sim⟩⟩ (2)




(119901 119902) isin [0 1]


t1

1 minus 120572 120572

tnminus1 tn

S(Δ) = 1








120593119899(119901 119902) = 120572 sdot 120593


(119901 119902) (3)









119899minus1to 119905119899 That is

120572 =

1 minus (119905119899minus1

minus 1199051

119905119899minus 1199051

)

2

if 119905119899gt 1199051

1 if 119905119899= 1199051

(4)









0 That is

120582119899(119901 119902) =

(119899 minus 1) 120582119899minus1

(119901 119902) + 120582119905now (119901 119902)

119899 (5)





119899(119901 119902) to denote



119899(119901 119902) is only 025 then




119899minus1(119901 119902)


0(119901 119902) = 0


120588 = 119890minus120581Δ119905

(6)




Time decay curve

k = 050

k = 010

k = 005

k = 001

100 200 300 400 5000Δt

0

01

02

03

04

05

06

076

08

09

1

Tim

e dec

ay v

alue








119903119894119895=

120593119899(119894 119895) if 119894 = 119895

1 if 119894 = 119895

(7)


119894119895= 0 Then

119877 = (119903119894119895)119899times119899

=

[[[[[

[

1 11990312


11990321



1199031198991

1199031198992

sdot sdot sdot 1

]]]]]

]

(8)




119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2


(120593119899(119902 119888) minus 120593

119899(119888))2

(9)






119901119902| reaches




119899(119901 119902) return




sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)



119899(119901 119902) = sim

119899(119902 119901)


p q

dT(p q)


qpiT(p q)







That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)







Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








3 IDTrust-Arch







Evidence collector

Trust decision

Original P2P system


Trust degree

Request




evaluation

EC(i) TD(i) P2P(i)

fen

fenminus1gnminus1

gn


Explicit evidence


Implicit evidence


Decay⟨120588⟩

Similarity⟨sim⟩

EC(i)




Trans value⟨⟩


TD(i)

P2P service


P2P(i) TD(i) EC(i)

Node i

(d) Topology







119899(1 2)



119891119890119899(119901 119902) = update (119891119890

119899minus1(119901 119902) 119892

119899(119901 119902))

119892119899(119901 119902) = update (119892

119899minus1(119901 119902) evidence (119901 119902))

(1)


119899(119901 119902)denotes





Evidence (119901 119902) = ⟨⟨120593⟩ ⟨120582 ] 120588 sim⟩⟩ (2)




(119901 119902) isin [0 1]


t1

1 minus 120572 120572

tnminus1 tn

S(Δ) = 1








120593119899(119901 119902) = 120572 sdot 120593


(119901 119902) (3)









119899minus1to 119905119899 That is

120572 =

1 minus (119905119899minus1

minus 1199051

119905119899minus 1199051

)

2

if 119905119899gt 1199051

1 if 119905119899= 1199051

(4)









0 That is

120582119899(119901 119902) =

(119899 minus 1) 120582119899minus1

(119901 119902) + 120582119905now (119901 119902)

119899 (5)





119899(119901 119902) to denote



119899(119901 119902) is only 025 then




119899minus1(119901 119902)


0(119901 119902) = 0


120588 = 119890minus120581Δ119905

(6)




Time decay curve

k = 050

k = 010

k = 005

k = 001

100 200 300 400 5000Δt

0

01

02

03

04

05

06

076

08

09

1

Tim

e dec

ay v

alue








119903119894119895=

120593119899(119894 119895) if 119894 = 119895

1 if 119894 = 119895

(7)


119894119895= 0 Then

119877 = (119903119894119895)119899times119899

=

[[[[[

[

1 11990312


11990321



1199031198991

1199031198992

sdot sdot sdot 1

]]]]]

]

(8)




119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2


(120593119899(119902 119888) minus 120593

119899(119888))2

(9)






119901119902| reaches




119899(119901 119902) return




sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)



119899(119901 119902) = sim

119899(119902 119901)


p q

dT(p q)


qpiT(p q)







That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)







Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Evidence collector

Trust decision

Original P2P system


Trust degree

Request




evaluation

EC(i) TD(i) P2P(i)

fen

fenminus1gnminus1

gn


Explicit evidence


Implicit evidence


Decay⟨120588⟩

Similarity⟨sim⟩

EC(i)




Trans value⟨⟩


TD(i)

P2P service


P2P(i) TD(i) EC(i)

Node i

(d) Topology







119899(1 2)



119891119890119899(119901 119902) = update (119891119890

119899minus1(119901 119902) 119892

119899(119901 119902))

119892119899(119901 119902) = update (119892

119899minus1(119901 119902) evidence (119901 119902))

(1)


119899(119901 119902)denotes





Evidence (119901 119902) = ⟨⟨120593⟩ ⟨120582 ] 120588 sim⟩⟩ (2)




(119901 119902) isin [0 1]


t1

1 minus 120572 120572

tnminus1 tn

S(Δ) = 1








120593119899(119901 119902) = 120572 sdot 120593


(119901 119902) (3)









119899minus1to 119905119899 That is

120572 =

1 minus (119905119899minus1

minus 1199051

119905119899minus 1199051

)

2

if 119905119899gt 1199051

1 if 119905119899= 1199051

(4)









0 That is

120582119899(119901 119902) =

(119899 minus 1) 120582119899minus1

(119901 119902) + 120582119905now (119901 119902)

119899 (5)





119899(119901 119902) to denote



119899(119901 119902) is only 025 then




119899minus1(119901 119902)


0(119901 119902) = 0


120588 = 119890minus120581Δ119905

(6)




Time decay curve

k = 050

k = 010

k = 005

k = 001

100 200 300 400 5000Δt

0

01

02

03

04

05

06

076

08

09

1

Tim

e dec

ay v

alue








119903119894119895=

120593119899(119894 119895) if 119894 = 119895

1 if 119894 = 119895

(7)


119894119895= 0 Then

119877 = (119903119894119895)119899times119899

=

[[[[[

[

1 11990312


11990321



1199031198991

1199031198992

sdot sdot sdot 1

]]]]]

]

(8)




119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2


(120593119899(119902 119888) minus 120593

119899(119888))2

(9)






119901119902| reaches




119899(119901 119902) return




sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)



119899(119901 119902) = sim

119899(119902 119901)


p q

dT(p q)


qpiT(p q)







That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)







Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






119899(1 2)



119891119890119899(119901 119902) = update (119891119890

119899minus1(119901 119902) 119892

119899(119901 119902))

119892119899(119901 119902) = update (119892

119899minus1(119901 119902) evidence (119901 119902))

(1)


119899(119901 119902)denotes





Evidence (119901 119902) = ⟨⟨120593⟩ ⟨120582 ] 120588 sim⟩⟩ (2)




(119901 119902) isin [0 1]


t1

1 minus 120572 120572

tnminus1 tn

S(Δ) = 1








120593119899(119901 119902) = 120572 sdot 120593


(119901 119902) (3)









119899minus1to 119905119899 That is

120572 =

1 minus (119905119899minus1

minus 1199051

119905119899minus 1199051

)

2

if 119905119899gt 1199051

1 if 119905119899= 1199051

(4)









0 That is

120582119899(119901 119902) =

(119899 minus 1) 120582119899minus1

(119901 119902) + 120582119905now (119901 119902)

119899 (5)





119899(119901 119902) to denote



119899(119901 119902) is only 025 then




119899minus1(119901 119902)


0(119901 119902) = 0


120588 = 119890minus120581Δ119905

(6)




Time decay curve

k = 050

k = 010

k = 005

k = 001

100 200 300 400 5000Δt

0

01

02

03

04

05

06

076

08

09

1

Tim

e dec

ay v

alue








119903119894119895=

120593119899(119894 119895) if 119894 = 119895

1 if 119894 = 119895

(7)


119894119895= 0 Then

119877 = (119903119894119895)119899times119899

=

[[[[[

[

1 11990312


11990321



1199031198991

1199031198992

sdot sdot sdot 1

]]]]]

]

(8)




119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2


(120593119899(119902 119888) minus 120593

119899(119888))2

(9)






119901119902| reaches




119899(119901 119902) return




sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)



119899(119901 119902) = sim

119899(119902 119901)


p q

dT(p q)


qpiT(p q)







That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)







Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








0 That is

120582119899(119901 119902) =

(119899 minus 1) 120582119899minus1

(119901 119902) + 120582119905now (119901 119902)

119899 (5)





119899(119901 119902) to denote



119899(119901 119902) is only 025 then




119899minus1(119901 119902)


0(119901 119902) = 0


120588 = 119890minus120581Δ119905

(6)




Time decay curve

k = 050

k = 010

k = 005

k = 001

100 200 300 400 5000Δt

0

01

02

03

04

05

06

076

08

09

1

Tim

e dec

ay v

alue








119903119894119895=

120593119899(119894 119895) if 119894 = 119895

1 if 119894 = 119895

(7)


119894119895= 0 Then

119877 = (119903119894119895)119899times119899

=

[[[[[

[

1 11990312


11990321



1199031198991

1199031198992

sdot sdot sdot 1

]]]]]

]

(8)




119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2


(120593119899(119902 119888) minus 120593

119899(119888))2

(9)






119901119902| reaches




119899(119901 119902) return




sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)



119899(119901 119902) = sim

119899(119902 119901)


p q

dT(p q)


qpiT(p q)







That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)







Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





119888 sim119899(119901 119902)

=

sum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888)) sdot (120593

119899(119902 119888) minus 120593

119899(119888))

radicsum119888isin119868119901119902

(120593119899(119901 119888) minus 120593

119899(119888))2


(120593119899(119902 119888) minus 120593

119899(119888))2

(9)






119901119902| reaches




119899(119901 119902) return




sim119899(119901 119902)

=

0 if (10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [minus1 0))

119888 sim119899(119901 119902) elseif (

10038161003816100381610038161003816119868119901119902

10038161003816100381610038161003816gt 120591) and (119888 sim

119899(119901 119902) isin [0 1])

05 otherwise

(10)



119899(119901 119902) = sim

119899(119902 119901)


p q

dT(p q)


qpiT(p q)







That is

dT119899(119901 119902) = 120593

119899(119901 119902) (11)







Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






Node p

ReturnSP list qSet

Rank decTList darr

For each q in qSet

Calculate

in qSet


No

Yes

Node q

Servicerequest




Is q the last one





=


|119888Set| if |119888Set| gt 0

0 otherwise

(12)



gT119899(119901 119902) = 120588 sdot (120577 sdot dT

119899(119901 119902) + (1 minus 120577) sdot iT

119899(119901 119902))


(119901 119902)

(13)



Let decT119899(119901 119902) = gT

119899(119901 119902)sdot]




value ]1010

(1 3) = 100 we can get decT1010

(1 3) = 65


119899(119901 119902) is a global













Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in









Trust decision (TD)

Request

Response


Request

Requ

est

Response

Resp

onse


Request

Response

Node y

Extension part

Node x












A nodeB nodeC node

75250


A nodeB nodeC node

75025


A nodeB nodeC node

701515










119901119902|







20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1G

loba

l tru

st de

gree




20 40 60 80 1000Iterations

0

01

02

03

04

05

06

07

08

09

1

Glo

bal t

rust

degr





0

0005

001

0015

002

0025

003

0035

Trus

t com

puta

tion

time (

s)




0

1

2

3

4

5

6

Trus

t com

puta

tion

time (

s)








0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




0005

015

025

035045

Tran

sact

ion

rate

with

mal

icio

us n

odes

(0-1



EigenRepRanTrust

01

02

0304

05

10 20 30 40 50 60 70 80 90 1000Iterations






(14)


7 Conclusions







Acknowledgments


References













































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in











































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



research article a novel iterative and dynamic trust...

Documents