Computing trust levels based on user’s personality and observed system trustworthiness

Abstract

In this article, we describe an approach for computing the current trust level of individual users towards an online system and present initial validation results from a small-scale experiment. This trust computational model relies upon survey research for identifying the set of key trust attributes and grouping users into four segments of expected behaviors. Each user’s initial trust level is computed based on a set of assumptions tailored to the specific segment she belongs to, while the trust level evolution takes additionally into account the system outcomes she has experienced so far. More specifically, the trust update follows a machine learning approach, where during the training phase that consists of a small number of system outcomes, users are asked to report their actual trust levels. Finally, we demonstrate the trustors’ segmentation validity and trust estimation accuracy by performing a small-scale experiment in the context of a fictitious online security service.