Title: PROFILING IN AMI DESIGN

Profiling in AmI design

Profiles: predefined, automated and other types of profiles

In the beginning stages of AmI it may be the case that end-users and AmI service providers work with predefined profiles. The service provider may target certain customers (for instance of a certain age) and the end-user may provide input via a deliberately constructed profile (set of preferences concerning a specific environment, for instance a restaurant, hotel, car). In that case AmI does not make use of profiling technologies in the sense of this Workpackage, and one can doubt seriously whether we can - in that case - speak of an intelligent environment, since the environment does not ‘learn’ (only applies predefined profiles). 

We shall distinguish this type of profile as ‘sets of data’ from the type that is derived from profiling technologies (the focus of this Workpackage) as ‘sets of correlated data’.  

In FIDIS deliverable 7.2 user modelling and user adaptive applications are discussed as a type of personalised profiling. To some extent this type of profiling does not depend only on data mining techniques, as it makes use of direct input by the end-user, simple extraction out of data bases (without use of stochastic data mining techniques), capture of user’s activities and inference from other type of information. User modelling seems a mix of predefined profiles, automated profiles and human intervention, thus being more elaborate and more sophisticated than data mining itself. It seems however that user modelling does result in a dynamic set of correlated data, which is used as a knowledge construct to create an adaptive environment, so the profiles that are generated during a process of user modelling do fall within the scope of profiles as sets of correlated data. 

The context of the human person

An AmI environment should - in order to function as proposed - not only profile the data subject whose environment will be customised, but also this subject’s context. To make the environment adaptive to the inferred preferences of the subject, the context itself will have to be profiled. This concerns data like room temperature, volume of the audio-set, amount of light and/or the presence of certain objects and even, to complicate matters, other subjects that have an equal ‘right’ to personalised services in the particular environment. This would mean that, as far as nonhumans are concerned, the definition of profiling should be rearticulated as: 

“The process of constructing profiles (correlated data) that identify either a data-subject or a group/category/cluster, and/or the application of profiles (correlated data) to a data-subject as a member of a specific group/category/cluster.”

As indicated in the glossary, a data subject is the subject (human or nonhuman) that the data refer to/describe/are attributed to. By using the term data subject the scope of profiling is widened to include any human, animal or thing of which data are processed and stored.