Background: The primary goals of a reputation management scheme are determining the service quality of the peers who provide a service (i.e., service providers) by using feedback from the peers who have rated the service (i.e., raters), and determining the trustworthiness of the raters by analyzing the feedback they provide about the service providers. Thus, the success of a reputation management scheme depends on the robustness of the mechanism to accurately evaluate the reputations of the service providers and the trustworthiness of the raters.
As in every security system, trust and reputation management systems are subject to malicious behaviors. Malicious raters may attack particular service providers (e.g., sellers) to undermine their reputations while helping other service providers by boosting their reputations. Malicious service providers may also provide good service qualities (or sell high-quality products) to certain customers in order to keep their reputations high while cheating other customers unlikely to provide feedback. Moreover, malicious raters or service providers may collaboratively mount sophisticated attack strategies by exploiting their prior knowledge about the reputation mechanism. Hence, building a resilient trust and reputation management system that is robust against malicious activities is a challenging issue.
Various systems exist for enabling reputation management between service providers and users, but the current systems lack several benefits of the present disclosed technology. There is a need for efficient, reliable, and scalable reputation management schemes that resist impact from user dishonesty and unreliability and are resilient to malicious attacks.
Technology: Faramarz Fekri and Arman Ayday from the School of Electrical and Computer Engineering at Georgia Tech have created the reputation management methods that can include receiving a plurality of ratings. Each rating can be associated with a service provider and a rater. The method can further include modeling the service providers, the raters, and the ratings as a factor graph representing the factorization of a joint probability distribution function of variables, calculating the marginal distributions using a belief propagation algorithm applied to the factor graph, and determining reputation values associated with the service providers and trustworthiness values associated with the raters based on the calculating. Each factor node of the factor graph can correspond to a rater and be associated with a local function representing marginal distributions of a subset of the variables. The subset of variables can include a trustworthiness value associated with the rater and one or more ratings associated with the rater. Each variable node of the factor graph can correspond to a service provider and each service provider can be associated with a reputation value.
Calculating can include iteratively passing messages between factor nodes and variable nodes connected by an edge. An edge can connect a factor node and a variable node and can represent one or more ratings associated with a rater and a service provider corresponding to the factor node and variable node, respectively. Messages can be passed until a termination condition is reached. The termination condition can be reached when the determined reputation values for one or more of the service providers remain constant between at least two iterations. Alternatively, the termination condition can be reached after a certain number of iterations. A message from a variable node to the factor node can represent a probability that a reputation value associated with the corresponding service provider equals a certain value at a current iteration. A message from a factor node to a variable node represents a probability that a reputation value associated with a corresponding service provider equals a certain value given one or more ratings between the corresponding service provider and a corresponding rater and the trustworthiness value associated with the corresponding rater at the current iteration.