Increasing users' trust on personal assistance software using a domain-neutral high-level user model

Abstract

People delegate tasks only if they trust the one that is going to execute them, who can be a person or a system. Current approaches mostly focus on creating methods (elicitation approaches or learning algorithms) that aim at increasing the accuracy of (internal) user models. However, the existence of a chance of a method giving a wrong answer decreases users' trust on software systems, thus preventing the task delegation. We aim at increasing users' trust on personal assistance software based on agents by exposing a high-level user model to users, which brings two main advantages: (i) users are able to understand and verify how the system is modeling them (transparency); and (ii) it empowers users to control and make adjustments on their agents. This paper focuses on describing a domain-neutral user metamodel, which allows instantiating high-level user models with configurations and preferences. In addition, we present a two-level software architecture that supports the development of systems with high-level user models and a mechanism that keeps this model consistent with the underlying implementation. © 2010 Springer-Verlag.