Evaluating agents' trustworthiness within virtual societies in case of no direct experience. (December 2020)

• Record Type: Journal Article
• Authors: Sapienza, Alessandro; Falcone, Rino
• Is Part Of: Cognitive systems research. Volume 64(2020)
• Page Start: 164
• Page End: 173
• Publication Date: 2020-12
• Subjects: Social recommendation -- Trust -- Multi-agent system; Cognition -- Periodicals; Cognitive engineering (System design) -- Periodicals; Artificial intelligence -- Periodicals; 153.05
• DOI: 10.1016/j.cogsys.2020.08.005
• Languages: English
• ISSNs: 1389-0417

Abstract

Abstract: A great deal of effort has been made to introduce trust models to assess trustworthiness within virtual societies. The great majority of them makes extensive use of direct experience as the main source of information, considering recommendation/reputation and inferential processes just later, as a secondary mechanism to refine trust assessment. In this kind of networks, unfortunately, direct experience might not always represent the best solution to assess trustworthiness. In fact, their highly dynamic structure promotes an increase of the average number of interconnections among agents. This in turn negatively affects the degree of knowledge the agents possess about each specific individual, i.e. direct experience. To date, however, it has not been said much about how to face these situations. It is fundamental to find an effective approach for trust assessment even in lack of direct experience, which is the central focus of this research. By the means of a multi-agent social simulation, we consider the situation in which an agent can just access indirect knowledge for trust assessment, namely recommendations of specific individuals or whole categories of individuals. Then, we compare the efficiency of these two approaches in order to identify when it is more convenient to rely on the first or on the second one. As expected, our results confirm that the dynamic nature of these networks strongly affects the role of categories. We modeled this feature introducing the "turnover" in the simulations, whereby the higher is the turnover the more convenient it is relying on categories. Besides this confirmatory result, our simulations highlight the higher degree of robustness of categories in the presence of unreliable recommenders. Such a result is even more significant if there is no available information about how reliable the recommenders are. The results we obtained are in accordance with the current literature and can be of important interest for the development of this sector.