Research On Trust And Reputation Model In Multi-Agent System

Agent and the Multi-Agent System(MAS) are important research topics in the artificial intelligence and computer science domains. Trust plays an important role in the interaction of human society, and is focused on in many research domains. Trust mechanism is also introduced to open MAS to help agents choose interaction partners, which not only is a very meaningful subject for research but also has application value. Agent is the abstract of interaction entity in computational world, so the trust study in MAS should take on responsibility of finding the general law of trust among compu-tational entities. And the general trust model that helps distributed agents evaluate the target need to be paid attention to.Inaccurate recommendation and interoperation between heterogeneous trust and rep-utation systems of MAS are studied in this dissertation. An application is also developed to research the development methodology. The research results of this dissertation are showed as follows:(a) The recommendation from witnesses may be inaccurate in MAS trust model. Inaccurate recommendation comes from the witness deliberate lying or from other factors independent of the witness. In the existing trust models, the method completely laying to witnesses is neither fair to witnesses, nor helpful to choose good interaction targets. In this dissertation, the operation radius of target agents is used to represent non-witness-side factors leading to inaccurate recommendation. In this way, the evaluator agent can revise the recommendation from its own viewpoint. And the algorithm to learn the operation radius of target can help evaluators gain greater utility. The experiments based on FIRE trust model show that the evaluator agents can select better targets in both static and dynamic environments with our method.(b) The grey system theory suits the situation with less information and uncertainty problems, and distributed MAS has the obvious grey characteristics. In our new trust model, GTrust, in order to guide the agent to utilize the recommendation from witnesses reasonably and avoid the negative impact from all kinds of inaccurate recommendation factors, including the deliberate lying, a mechanism is designed to rate the witnesses' recommendation action, grey sequence generation method is used to fill the incomplete data, and grey fixed weight clustering approach is used to determine the trust to witnesses. Simulation results show that the strategy in GTrust is a practical choice for the complex environment, and GTrust can help evaluators achieve better interaction results.(c) A platform is designed to support interoperation among heterogeneous reputation models, based on Functional Ontology of Reputation (FORe) and the existing reputation models. The dissertation formally defined this platform, including the basic elements of agent, rating, recommendation, and so on. The reputation of target and that of wit-ness are distinguished in the platform and four forms of recommendation information are proposed to support the existing and future models. This work may be beneficial to com-bine heterogeneous reputation models, to integrate applications with different reputation models and to measure the performance of reputation models.(d) Although some literature management softwares can help researchers get liter-atures, there are still some limitations, such as lack of cooperation in search, monotony literature sources and deficient personality, and so on. ALRS, an agent-based literature recommendation system, is developed to help researchers cooperate when searching and sharing literatures. Agents in ALRS, which play both roles of a searcher and a recom-mender, mimic human interactions and enhance the sources of literature. The well-chosen interaction protocol and decision method based on accumulated experiences make it pos-sible for agents to choose the right recommender to provide literatures. On the basis of accumulated experience, the trust is inferred to enhance the effect of decision making in the stage of selecting the interactive partners.