Reputation systems are nowadays widely used to support decision making in networked systems. Parties in such systems rate each other and use shared ratings to compute reputation scores that drive their interactions. The existence of many reputation systems with remarkable differences calls for software frameworks for describing, implementing and comparing their performances while taking into account the architecture of the network where the systems have to be deployed. To tackle this issue we propose a software framework for network-aware evaluations of reputation systems based on the notion of probabilistic trust. This tool enables rapid prototyping and evaluation of reputation system models, by taking explicitly into account the networked execution environment. To implement specific models we just enrich Klava (a network-aware extension of Java) with additional classes. To assess performances of the resulting reputation systems, the implementation is analysed by a control manager that performs experiments according to user-specified parameters. The developed framework relies on the formal foundations of Klaim, a network-aware coordination language, and its Java implementation Klava. Feasibility and effectiveness of our proposal is demonstrated by reporting on the analysis of two simple models of reputation systems.
|
