Enhancing Trust-Based Competitive Multi Agent Systems by Certified Reputation

Trust models play a crucial role in many application contexts involving competition like e-commerce, e-learning, etc. The above application domains deal with very-large-size sets of users with heterogeneous platforms. As a consequence, the Multi-Agent Systems paradigm appears to be one of the most promising approaches to apply in this context. Verifying the trustworthiness of a reputation feedback (recommendation) provided by an agent is a crucial issue to face when designing reputation models for competitive Multi-Agent Systems. In the past, the experience of the ART community highlighted that, in absence of information about the quality of the recommendation providers, it is better to exploit only the direct knowledge about the environment (i.e., a reliability measure), missing the reputation measure. However, when the size of the agent space becomes large enough, and the number of “expert” agents to contact is small, the use of just the reliability is little effective. Unfortunately, the largeness of the agent space makes the problem of the trustworthiness of recommendations very critical, so that the combination of reliability and reputation is not a trivial task. In this paper, we deal with the above problem by studying how the introduction of the notion of certified reputation, and its exploitation for combining reputation and reliability, can improve the performance of an agent in a competitive MAS context. We analyze different populations, using the standard platform ART, highlighting a significant positive impact and providing very interesting results.