Reputation plays an important role in multi-agent system. It is a socialized form of trust which makes agent cooperate with each other and reduces the cost of agents' interaction. In a world with only computational trust, the agent can only perceive its own interactions. Its learned trust pattern can only be used by itself. There is no socialized mechanism to magnify the trustworthiness that has been learned. To introduce reputation is the solution to efficiently exploit the trust patterns. If the NTR algorithm is designed for intelligent agents, then the reputation propagation models and reputation generation mechanism are designed for multi-agent systems. Introducing reputation into multi-agent systems brings many benefits: the agent can greatly extend its range of influence to cover other agents. The agent also can share the interaction experience with others. Such sharing will accelerate the washing out of malevolent agents and increase the possibility of transactions for benevolent agents. The reputation will improve the executive efficiency of agents by avoiding unnecessary communication and transactions. In general, reputation is the key to form a tight coupling agent society. There is no acknowledged or standard definition for computational reputation. But it is possible to describe it from five facets: interaction experience, intention of propagation, range of propagation, path of propagation, content of reputation. Interaction experience explains the reputation from the view of information source; intention of propagation explains from the view of agents' motivation; range of propagation explains from the view of spatial consideration; path of propagation explains from the view of network; content of reputation explains from the expression of the reputation. The author builds three models of reputation propagation. Point-to-point based inquiry allows an initiative agent start an inquiry request to its acquaintance. If the middle agent has intention to transfer the inquiry, then the request can be propagated far from the initiative agent and thus form a reputation network. Broadcasting based propagation is to let agent broadcast its experience about every interaction or transaction so that every other agents in the society can learn what happened.
TopDefinition
The meaning of reputation in dictionary implicitly refers to the reputation in the human society. There is a big difference between the human reputation and computer based reputation. In addition to the common meaning of the word “reputation” appeared in the dictionary, the author tries to depict the computational reputation, that is, agent reputation from five aspects: interaction experience, intention of propagation, range of propagation, path of propagation, content of propagation. Figure 1 illustrates these five aspects.
Interaction experience toward target agent is the source of reputation. An agent first needs to have some interaction experience with the specific target agent. And such reputation interaction needs to have some results either positive or negative. As the research describes in the previous section, the agent perceives the result and melts it with its existing pattern toward the target agent. In case of inquiry on the target reputation or in case of broadcasting, the agent will generate a reputation according to the trust pattern and propagate the reputation to some agents or some networks. Trust recognition is the very first step of reputation propagation, only when agents have interacted with some other agents, then they are qualified with the “experience”, pattern of trustworthiness, to tell the other agents that agent x is good and agent y is bad.
Figure 1.
Definition of reputation: the dimension of defining a computational
Intention of propagation is the intent that an agent is willing to propagate its perceived reputation to the other agents. It is like a switch that controls whether the agent is willing to share its interaction experience with its fellow agents over some form of reputation networks. The set of the switch can be used with the goal of an agent to emphasize that an agent is proactive and on its own initiative to participate a network. In this research, the author assumes every agent has the initiative to share their interaction experience because the purpose of the research is to investigate the mechanism of trust and reputation in a dense network full of active sharing agents.
Range of propagation is the distance that an agent can affect another agent by sending its perceived reputation to. The range here refers to the number of agents that lies between the sending agent and the receiving agent. In a virtual world of computer program, the simplest scenario is to remove the limitation of propagation range, let reputation propagated without any blocking or decaying. Any agent can get undistorted reputation from the remote agent. Agents in such scenario are actually stay in a flat and homogeneous network without layers or clusters which is too nave from the real world networks. In the real world networks, the network is layered or clustered into LAN, MAN or WAN, accordingly, the agents are belong to different owner in different application domain with different goals and usages. In the complex scenario, it is not reasonable and not appropriate to propagate the reputation unchangeable. The reputation could be enhanced or decayed according to the trustworthiness between transmitting agents or the agent network structures.
Propagation Mechanism Design
Through the description in section 3, it is clear that designing ways of propagation is a critical task for spreading recognized reputation. The effectiveness of a propagation method can be analyzed from three perspectives: coverage, performance and structure. Wide coverage means that the reputation message should be propagated to as many agents as possible and as far distance as possible. High performance means that the propagation should efficiently make use of the network; the reputation message should reach the agents quickly without any distortion. Simple structure means the propagation mechanism should easy to implement and have a clear layered structure. According to those three principles, the research designs three forms of reputation propagation mechanisms: point-to-point based inquiry, broadcasting based propagation and observer based propagation. The next three subsections describe how those three mechanisms are designed, their detailed propagating processes, and their advantages and disadvantages.