A Model for Experience-Based Agent Specific Trust

Contemporary computerized tasks increasingly depend on inherently inaccurate information provided by autonomous agents. Reasons for the information inaccuracy are many, including the uncertainty in measurement (calibration errors), the process inherent inaccuracy (rounding), changing level of quality and, when humans are involved, differences in preferences (bias), (un)intentional violation of expectation to mention a few. Parameterization of this inaccuracy is demanded for prompt and justified adaption. Frequently, this parameterization is overlooked when models for reasoning on the inaccuracies are presented. In this paper, we address the parameterization of inaccuracy by an experience-based model. The model is based on Dempster-Shafer theory of evidence that relies on a history of experiences of subjective satisfaction on some provided data. From this history, the model derives the warranted belief and certainty that may justifiably be placed on the acquired data. The model facilitates decay and abstraction of a subset of history to a versatile score. This paper’s contribution is in showing the experience-based model’s generality and versatility by mapping it to EigenTrust, Subjective Logic and probabilistic trust management models.