The role of trust measures is particularly relevant in competitive multi-agent systems. Recent studies highlight the importance of correctly balancing direct measures, as the reliability, and indirect measures, as the reputation. The key problem is that an agent may have an insufficient direct knowledge of another agent, showing the necessity of using a reputation measure, computed using some “gossip” coming from other agents of the community. However, in a community consisting of competitive agents, an agent can show a fraudulent or misleading behaviour, to create problems to its competitors. Consequently when an agent has to select the most promising interlocutors, it should be capable to assign a suitable weight to the reputation with respect to the reliability. This weight strictly depends on the number of interactions that the two agents have executed in the past (a high value implies to give a preference to the reliability), as well as on other information that the agent can acquire about the agent community during its life. In this paper, we introduce a trust model for a competitive agent, that considers both the aforementioned issues to combine reliability and reputation, dynamically adapting the coefficient that represents the percentage of importance the agent assigns to the reliability with respect to the reputation. We have evaluated this model on the well-known ART platform, clearly showing that this dynamic adaptation leads to improve the agent performances with respect to use a static value for the coefficient above.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.