Learning user preferences for adaptive pervasive environments: An incremental and temporal approach

Abstract

Personalization mechanisms often employ behavior monitoring and machine learning techniques to aid the user in the creation and management of a preference set that is used to drive the adaptation of environments and resources in line with individual user needs. This article reviews several of the personalization solutions provided to date and proposes two hypotheses: (A) an incremental machine learning approach is better suited to the preference learning problem as opposed to the commonly employed batch learning techniques, (B) temporal data related to the duration that user context states and preference settings endure is a beneficial input to a preference learning solution. These two hypotheses are the cornerstones of the Dynamic Incremental Associative Neural NEtwork (DIANNE) developed as a tailored solution to preference learning in a pervasive environment. DIANNE has been evaluated in two ways: first, by applying it to benchmark datasets to test DIANNE's performance and scalability as a machine learning solution; second, by end-users in live trials to determine the validity of the proposed hypotheses and to evaluate DIANNE's utility as a preference learning solution.