Learning preferences for personalisation in a pervasive environment

Abstract

With ever increasing accessibility to technological devices, services and applications there is also an increasing burden on the end user to manage and configure such resources. This burden will continue to increase as the vision of pervasive environments, with ubiquitous access to a plethora of resources, continues to become a reality. It is key that appropriate mechanisms to relieve the user of such burdens are developed and provided. These mechanisms include personalisation systems that can adapt resources on behalf of the user in an appropriate way based on the user's current context and goals. The key knowledge base of many personalisation systems is the set of user preferences that indicate what adaptations should be performed under which contextual situations. This thesis investigates the challenges of developing a system that can learn such preferences by monitoring user behaviour within a pervasive environment. Based on the findings of related works and experience from EU project research, several key design requirements for such a system are identified. These requirements are used to drive the design of a system that can learn accurate and up to date preferences for personalisation in a pervasive environment. A standalone prototype of the preference learning system has been developed. In addition the preference learning system has been integrated into a pervasive platform developed through an EU research project. The preference learning system is fully evaluated in terms of its machine learning performance and also its utility in a pervasive environment with real end users.