Degree Name

Master of Computer Science - Research


School of Computer Science and Software Engineering


In the past twenty years, the multi-agent technology has been widely employed for developing agent-based systems. Currently, agent-based service-oriented applications have been widely applied in many complex domains such as e-markets, grid computing, e-governments and service-oriented software systems, across Internet and organizations. In this kind of service-oriented multi-agent systems, service providers (agents) and service consumers (agents) are autonomous entities and can enter and leave environments freely. How to select the most suitable service providers according to the requested services from consumers in such an open and dynamic environment is a very challenging issue.

The objectives of this thesis include (1) studying the challenging issues of trustbased service provider selection, (2) investigating the current approaches of trust models for service provider selection in general service-oriented multi-agent systems, and (3) developing new solutions for service provider selection to overcome several limitations in current approaches.

In this thesis, two trust models are proposed and developed. One is a Priority-based Trust (PBTrust) model for single service provider selection. The other is a Group Service Trust (GTrust) model for group service providers selection when a complex service requests multiple service providers.

The designing purpose of the PBTrust model is to help service consumers in multiagent systems to select the most suitable single service providers. To deal with the provider selection problem, firstly the PBTrust model uses a rich context service description to represent service requests by confederating different attributes of a service and uses priority values to distinguish the importance of these attributes. This feature allows more objective evaluations on both required services and providers’ reputations. Moreover, the PBTrust model uses a relatively easy way to describe the different attributes of a service. Finally, the PBTrust model introduces the concept of experience weight which can avoid subjective and cheating references.

Being different with the PBTrust model, the GTrust model is designed for group service providers selection in service-oriented environments. Currently, many complex services are hard for single providers to fulfill the requests. Therefore, several service providers need to form groups to conduct the services. Developing trust model for group service providers selection is a hard topic, due to the structure of services composition and dependency relationships among services owned by different providers, the reputations of individual services and impacts of individual services on group performance in terms of their trust values towards to the group trust evaluation. The GTrust model offers several innovated mechanisms to help a consumer accurately evaluating the trust value for a group of services by taking the above features into account during group service providers selection. The GTrust model evaluates the trust value for a group of services by considering (1) the coverage rate of the requested functionalities from a service group, (2) the dependency relationships among individual services in a group, (3) the reference reports from third parties for each provider of individual services in a group and (4) the similarity measurement about to what extent the reference reports can reflect the new service request in terms of priority distributions on attributes of the requested service.