Abstract
In Multi-agent Systems, there are complex problems that cannot be solved by a single agent. Therefore, agent groups are formed to deal with the problems more effectively. These groups can have various types, structures, and complex behaviours. Meanwhile, the openness characteristic of multi-agent systems frequently introduces uncertainty into the environments. In such situations, it is hard or impossible to establish trust between agents and it can greatly affect the operation of the systems. Several trust models have been proposed to assist agents to make decisions during moments of uncertainty. Nevertheless, most of the existing works focus on trust evaluation for individual agents. Since complex agent groups are becoming ubiquitous in agent societies, these trust models have been exposing several shortcomings in addressing different aspects of trust in agent groups. This includes the highly dynamic and complex behaviours of them. Therefore, it is necessary to have a comprehensive trust management mechanism regarding these groups.
To this end, our thesis presents a trust management stack which consists of three components, i.e., CoTrust, DBATE, and GEM. The three components manage trust in agent groups in three stages, i.e., group formation, group performance, and group disbandment. The following contributions have been made.
It provides an efficient trust establishment method regarding group formation: CoTrust is a trust establishment method devised to address the lack of effective trust mechanism in group formation. It includes a protocol and a preference reasoning method that helps agents to improve the success rate of cooperation requests and also the satisfaction of the cooperation.
It enhances the robustness of existing trust/reputation systems in MASs: GEM is a group evidence management method to address the inconsistency of evidence generated by the rating convention when the systems consist of agent groups. GEM provides an efficient rating distribution method for group members. By using an evidence-based approach, GEM can not only guarantee the fairness of the distribution but can also significantly improve the robustness of reputation systems.
It enables agent group trust evaluation in MASs: DBATE is a trust evaluation method which aims to address shortcomings of current trust evaluation models targeting ad-hoc groups. The approach combines the power of the dynamic Bayesian network and contextual information for improving the accuracy of the evaluation. All of the proposed methods come with empirical analyses to confirm their effectiveness.
