Jiminy: A Scalable Incentive-Based Architecture for Improving Rating Quality

Abstract

Abstract. In this paper we present the design, implementation, and evaluation of Jiminy: a framework for explicitly rewarding users who participate in reputation management systems by submitting ratings. To defend against participants who submit random or malicious ratings in order to accumulate rewards, Jiminy facilitates a probabilistic mechanism to detect dishonesty and halt rewards accordingly.Jiminy's reward model and honesty detection algorithm are presented and its cluster-based implementation is described. The proposed framework is evaluated using a large sample of real-world user ratings in order to demonstrate its effectiveness. Jiminy's performance and scalability are analysed through experimental evaluation. The system is shown to scale linearly with the on-demand addition of slave machines to the Jiminy cluster, allowing it to successfully process large problem spaces.

Towards reputation measurement in online social networks

“…To sociologists, reputations are aggregate assessments of firms' performance relative to expectations and norms in an institutional field [3]. From computer science's view, rating reputation is either based on scoring the entity as input, like in the reputation systems applied for e-commerce [4], or based on statistics derived from shared attitudes and opinions among members in the community [5].…”

A Robust and Knot-Aware Trust-Based Reputation Model

“…In Jimminy [12], the authors present an honesty assessment algorithm and a reward model for encouraging honest ratings. Their computation of honesty is based on the probability distribution of all ratings available for a subject.…”

“…Most of them compile an aggregated "general opinion" of all recommending members. Some of these systems disregard ratings that are too far from the popular rating score or even consider them as malicious [12]. We maintain that it is possible for the same experience to be perceived differently by different people.…”

Credible reputation systems for P2P e-communities

“…The reputation model proposed by Fernandes et al [94] and further enhanced by Kotsovinos et al [121], incorporates both rewards and punishment depending on the participant's behavior. More specifically, a credit balance is set up for every participant, which is credited with a reward for each opinion and debited for each recommendation query made by that user.…”

“…by mistake, and thus decrease system utility. Mechanisms for monitoring recent behavior and rewarding it if it is honest and keeps being honest for a certain period of time (as in [94], [121]) could help reputation restoration of previously misbehaving peers.…”