Privacy of trust in similarity estimation through secure computations

Abstract

Current e-commerce applications provide a binary definition of trust depending on whether requirements are satisfied or not. However, trust can be seen as a more general and subjective concept tightly linked to reputation. The subjectivity of the evaluations of certain behaviours justifies this kind of trust. We intend agents to manage subjective trust as humans do, in a vague and abstract way, as a fuzzy concept. In this paper we extend our previous work about the introduction of newcomers into an already running agent system. Because of the economical and strategic value of the trust-related information exchanged in such e-commercial scenarios, access to this information should be restricted to a certain group of close agents. Agents were grouped in cooperative clusters according to similarity of opinions about others. Specifically this paper evaluates the privacy of these opinions while similarity can be easily estimated. The results of this evaluation will justify the use of secure computations in the introduction of newcomers.