Trust Management In Multi-agent Systems

Along with the rapid development of Internet, distributed computing based on Internet is also developing rapidly and a lot of widespread and open systems have appeared. Multi-agent system (MAS) provides a new way to analyze, design and realize distributed and open systems. It is hailed as another significant breakthrough of software development. One of its most important characteristics is open. In MAS: (1) Agent may be selfish and it may mot completely accomplish its task; (2) Agent is impossible to hold all the knowledge about environment; (3) there isn't a central authority to control all the agents. According to the uncertainty above, trust plays an important role in agent interactions.Currently, research on trust management in multi-agent systems focuses on evidence-based and credential-based trust management. Evidence-based trust management (namely trust model) takes subjectivity, uncertainty and dynamics of trust into account. It forms trust relationships between agents by the way of evaluating their interaction history. Trust model presents the concepts of trust measurement, trust deduction, and trust evaluation, so that it can reflect the changeness and uncertainty of multi-agent systems finely. However, there isn't a universal trust model in multi-agent systems to dynamic handle trust relationships between agents. Currently, most of the models are either too simple to evaluate direct trust or combining multi trust path using arithmetically averaging method. Otherwise, they have no consideration of agents'cheapness.In credential-based trust management systems, trust relationship is achieved by credentials or credential chain, which represent the propagation of trust between principals. If a credential chain is not discovered, the trust relationship does not exist and access cannot be granted. If the trust relationship exists, a principal agent is fully trust another principal agent without question, revoking credentials will cancel trust relationship. The characteristic of such trust management is that it measures security too absolutely and does not consider the risk or uncertainty of trust, so that it can't satisfy the changeness of open environments.To deal with the problems exist in trust management of multi-agent systems, this thesis presents the D-S theory-based trust model FIRE+ and trust model-based automated trust negotiation.The main research contents of this thesis are as follows:(1) D-S theory based trust model FIRE+By combing the advantages of FIRE model and the model presented by Yu, this thesis presents the general trust model FIRE+. It introduces the presentation of trust information and the calculation method of each component in this model. Its characters as follows: With the representation of matrix, it efficiently increases the information used to evaluate trust value; Applying D-S theory and its combinational rules to evaluate and combine trust, so that uncertain information can been handled finely; Meanwhile, our approach adopts the concept of information-amount to derive each issue's weight and present the method of how to aggregate the whole trust value of target agent; In addition, it takes into account witness reliability.Moreover, it also discusses its application in web services selection. Based on the introduction of QoS definition, QoS characteristic of web services and multi- QoS parameters of web services, this thesis presents QoS-aware web services selection based on FIRE+ trust model. It effectively solves the problem of selecting web services blindly and makes the granularity of web services selection more finely. Experiments show that the new trust model presented by this thesis is more secure and effective that others.(2) Trust model-based automated trust negotiationBased on the traditional automated trust negotiation, this thesis takes subjective trust into account and develops the framework of trust model based automated trust negotiation. It presents the main components in the framework and their functions in detail. Trust model based automated trust negotiation brings about new research problems such as: trust model-based access policy, trust model-based negotiation strategy, trust model-based negotiation protocol and so on.This thesis emphasizes on the trust representation in access control policy and the negotiation processes of different negotiation strategies based on trust model. It presents trust model-based eager strategy and trust model-based parsimonious strategy. Also, the negotiation processes of these strategies are given. Moreover, it applies an application instance to show the negotiation process of trust model-based eager strategy.