A trust-based approach to clustering agents on the basis of their expertise

Abstract

The issue of detecting clusters of agents on the basis of their expertise in providing e-services is a crucial point in managing multiagent systems, since generally a cost is associated with a given expertise level and user would have the possibility to select those agents that have the best quality of service compatibly with his budget. However, estimating agents’ expertise in a multi-agent system is a hard task due to the generally large dimension of the system, so that only distributed approaches appear practicable, involving the participation of all the agents in the clustering task. In this paper, we propose to use the notion of trust as a basis for detecting clusters in a distributed manner. Our idea is based on the introduction of a simple trust model in a competitive multi-agent system, and on the assumption that the trust measure associated with each agent estimates the agent’s expertise, provided that a sufficient number of competition steps is performed. Some preliminary tests we have performed on the ART platform show that our approach provides good results with a limited number of clusters, while the clustering capabilities worse when the number of clusters increases.