In agent-based systems, agents can be organized within groups, called communities, where mem-bers are providing similar or complementary services. An example of such systems is agent-based communities of web services, where web services are abstracted as rational agents and empowered with decision making capabilities and can interact with each other. Managing reputation of each agent and of the whole community is a key issue towards securing this type of systems, where a con-troller agent is designed to observe and check the behavior of each member to update and maintainudthe system’s reputation. Scheduling the check (i.e. maintenance) by deciding about the moments where the check has to be done is still an open problem. Because it is highly expensive, maintenance cannot be done every moment or based on small history of agents’ behaviors. We propose in this thesis a scheduling algorithm that helps the controller agent improve the quality of the reputation mechanism, which increases the trust value of users toward the community. The proposed algorithm is based on a class of games called Bayesian Stackelberg. Our Bayesian Stackelberg game is designed between the controller agent and community members. We simulate and compare the efficiency of our algorithm with other stochastic techniques, namely uniform, normal and Poisson distributions. This research draws the lines for future work in the subject of optimizing reputation mechanisms through maintenance in different time intervals.
展开▼
