Plug and play context-awareness for pervasive environments

Sammanfattning: Pervasive environments are physical environments interwoven with a rich set of embedded sensors, actuators and displays offering their services to the user. The considerable amount of computing technology in such environments should be as unobtrusive as possible, aiding the user in everyday activities instead of being distracting. Context-awareness, i.e. the adaptiveness of the environment to the current situation of the user, is of tremendous importance for achieving the desired unobtrusiveness.This licentiate dissertation begins by looking at one aspect of pervasive systems where user experience can be considerably improved through context-awareness: of the thousands of services that may be available in a pervasive environment, only a small percentage will be of interest to the user in any given situation. Instead of having to issue explicit service discovery requests, users would like to be continuously informed about services relevant to them, depending on their current context and their preferences. In this dissertation a proactive service discovery approach for pervasive environments is proposed that addresses these implicit discovery requests. Services and user preferences are described by a formal context model which can effectively capture the dynamics of context and the relationship between services and context. Based on this model a set of algorithms is described that can continuously identify the services most relevant to the user in near real time.Proactive service discovery presumes detailed descriptions of user preferences and service capabilities. Manually providing this information is tedious and error-prone, hindering the adoption of the technology by non-expert users. In the second part of this dissertation it is shown how user preferences can instead be automatically identified by studying the user's interaction with the system. Furthermore an approach is presented for automatically providing service descriptions by observing the conditions under which services are executed and the typical effects they induce in the environment. By applying machine-learning techniques such as subspace clustering and classification on the observed data, these approaches are able to produce high-quality descriptions for both user preferences and service capabilities, without any need for manual annotation. Based on these results it is shown how discovery and learning algorithms can be seamlessly integrated into existing pervasive environments to provide plug and play context-awareness.