A General Model for Trust

A General Model for Trust

A GENERAL MODEL FOR TRUST

Abstract:

Reputation-based trust systems provide important capability in open and service-oriented computing environments. Mostexisting trust models fail to assess the variance of a reputation prediction. Moreover, the summation method, widely used for reputationfeedback aggregation, is vulnerable to malicious feedbacks. This paper presents a general trust model, called RLM, for a morecomprehensive and robust reputation evaluation. Concretely, we define a comprehensive reputation evaluation method based on twoattributes: reputation value and reputation prediction variance. The reputation predication variance serves as a quality measure of thereputation value computed based on aggregation of feedbacks. For feedback aggregation, we propose the novel Kalman aggregationmethod, which can inherently support robust trust evaluation. To defend against malicious and coordinated feedbacks, we design theExpectation Maximization algorithm to autonomously mitigate the influence of a malicious feedback, and further apply the hypothesistest method to resist malicious feedbacks precisely. Through theoreticalanalysis, we demonstrate the robustness of the RLM designagainst adulating and defaming attacks, two popular types of feedback attacks. Our experiments show that the RLM model caneffectively capture the reputation’s evolution and outperform the popular summation-based trust models in terms of both accuracy andattack resilience. Concretely, under the attack of collusive malicious feedbacks, RLM offers higher robustness for the reputationprediction and a lower false positive rate for the malicious feedback detection.

Existing System:

THE rapid growth of Internet and ubiquitous connectivityhas spurred the development of various collaborativecomputing systems such as service-oriented computing(SOC), Peer-to-Peer (P2P), and online community systems.In these applications, the service consumer usually knowslittle about the service providers, which often makes theconsumer accept the risk of working with some providerswithout prior interaction or experience. To mitigate thepotential risks of the consumers, reputation-based trustsystems are deployed as a popular approach topredict how much the service provider can be trusted. The reputation value plays a pivotal role in aggregating,filtering, and ordering information for consumers to selectservice providers, and it can act as an incentive for serviceproviders to improve their Quality-of-Service. Over the pastfew years, many reputation (social trust) models have beenproposed for different applications such as: social webservices decentralized overlay networks and applications, multiagent systems andrecommender systems.

Drawback:

1)Direct request for using service..

2)Unknown about service.

3)Unknown about provider.

4)Unable to collect feedback.

Proposed System:

Reputation is a statistical value about the trust probabilityderived from the behavior history. Usually, the reputation isbased on the interactions carried out directly betweenproviders and the evaluator (personal experience) and therecommendations made by other consumers (feedback).

From the personal experience’s perspective, most existingwork used the simple average, the Bayesian orthe belief models to quantify the trust as somestatistical values. However, they ignore another importantattribute of the predicted statistical value, namely theprediction variance (or prediction accuracy), which depictshow much the trust prediction may deviate from the real one.For example, a service provider has a service successprobability of 0.9. But due to the incomplete personalexperience, a customer quantifies the provider’s trust as0.7. By using existing trust models, the customer can neitherassess the accuracy of the reputation prediction made by hernor assess the trust values recommended by other in order touse them in her recommendation. Hence, it is hard for aconsumer to decide how much to rely on the prediction of thefeedbacks made by others to make her own trust decision.Moreover, when the customer recommends this trustprediction to others as a feedback, she cannot give reliablesuggestion about how to aggregate the feedback so thatothers can minimize the variance of their trust evaluation.

To aggregate feedbacks recommended by others, thesummation method is widely applied in reputation systems,such as eBay and Eigen trust. However,several have shown that it is easy to manipulate summation-based feedback aggregation by malicious nodes for their personal profits. A malicious node can falselyimprove its own reputation or degrade the reputations ofothers. As a measure to defend malicious feedbacks for thesummation method, most existing work weighted thefeedbacks by considering their credibility, such as the trustvalue based credibility used in Eigentrust and thepersonalized similarity (PSM)-based credibility used inPeer Trust. However, these credibility techniques usuallyneed accurate trust knowledge of the system or manually tuned intuitive parameters, which areoften unrealistic or impractical in a real world application.We believe that the feedback credibility based techniqueslack of the robustness to resist malicious feedbacks.

Modules & Description:

1)Comprehensive Trust Formulation:

We argue that to get a comprehensive trust prediction, trustmodels need to provide the local assessment of trustprediction accuracy. Since the reputation value is essentiallya statistical value derived from the observation samples(reputation feedbacks), we model the reputation in astatistical form. Assuming that the real reputation of anode is denoted by R, which is not known by the trustevaluators. In a comprehensive model, we try to predict theactual reputation value by trust evaluation, denoted as atwo dimension tuple, namely rep ¼ fhRi; Pg, where hRi isthe predicted reputation value, and P is the estimatedreputation prediction variance, which is estimationabout the square error between hRi and the real reputationR. The attribute P is an evaluation about the accuracy of thepredicted reputation value hRi, which can be understood asthe evaluator’s confidence in hRi. Hence, the lower theestimated prediction variance P is, the more confidence willthe evaluator have in the predicted reputation value hRi.Upon obtaining the prediction hRi and its estimatedprediction variance P for a node, the node can send thetuple rep to others as a reputation feedback. Hence, afeedback can be denoted as f ¼ fz; cg, where z (comingfrom hRi) is the feedback reputation value, and c (comingfrom P) is the suggested feedback variance, which indicateshow accurate the feedback reputation value z is, and servesas a hint to others about how to intelligently aggregate thefeedback reputation value. A bigger suggested feedbackvariance c means that the recommender has less confidencein z. Hence, the aggregator should reduce the influence ofthe feedback in his reputation aggregation.We dedicate Section 4 to the comprehensive trust evaluationproblem. Concretely, we use a linear hidden Markovprocess to track the evolution of trust state, and propose theKalman aggregation method for feedback aggregation insteadof using the intuitive summation method.

2)Malicious Feedback Attack Model:

Attackers in a reputation system can either work alone orlaunch attacks by colluding with one another. A collusiveattack can be implemented by disparate attackers or a singleattacker acquiring multiple identities through a Sybil attack. Typically, the effect of a single attacker is relativelysmall, but collusive attackers usually have much more severeinfluence on the reputation system. They can ooperate toissue high volumes of malicious feedbacks, which are moredifficult to defend against. Hence in this paper, we areprimarily concerned with the collusive reputation attackwhich has large number of malicious feedbacks.In this paper, we can classify the malicious feedback intotwo types: adulating feedbacks and defaming feedbacks. Inadulating feedback attacks, attackers try to falsely improvethe reputation of their own or their partners. One basicform of the attack occurs when the multiple colluding

Attackers send unfairly positive feedbacks about each other.The adulating feedback reputation value can be modeled intwo ways:

1. Random positive feedback: the feedback reputationvalue is a random valuebetween 1 and the predicted reputation value set bythe attacker. In such attacks, colluding attackers sendrandom feedback reputation values about the targetseparately without coherence.

2. Coordinated positive feedback: the feedback reputationvalue has a deterministic relationship with thedesired reputation value predicted by the attacker.In such attacks, all the participating attackers seek tosend feedback values coherent to each other. Considerthe example shown in Fig. 1b, given the estimated reputation value x, the reputation valuefor the coordinated positive feedback is ðx2 þ 1Þ=2

Software Requirements

Operating System:Windows XP/2003 or Linux/Solaris

User Interface:HTML, CSS

Client-side Scripting:JavaScript

Programming Language:PHP

Database:MySQL

Hardware Requirements

Processor:Pentium IV

Hard Disk:40GB

RAM:256MB

References

(1) PHP and MySQL Web Development by Luke Welling and Laura Thomson

(2) PHP5 and MySQL by W Jason Gilmore (Apress)

(3) PHP Web Development by Allan Kent and David Powers (Apress)

(4) Wiki pedía, URL:

(5)

(6) Google, URL:

Technologies to be used

HTML, CSS (Web Presentation )

JavaScript (Client-side Scripting)

PHP (as programming language)

MySQL (Database)

Windows XP/2003 or Fedora (Operating Systems)

Apache, XAMPP

Contact: 9030211322, 9030411292

Email id: ,