Title: TECHNICAL ASPECTS OF PROFILING IN AMI

Technical aspects of Profiling in AmI

To fulfil the full AmI vision, it is proposed that the AmI acts according to the user’s preferences, needs and expectations, thus profiling is the key stone of this scenario. From an implementation perspective, the ‘intelligent agent’ is the embodiment of the profiling aspect which attempts to build a comprehensive profile of the user by processing data recorded from his or her interactions, behaviour, preferences, and essentially ‘learning’ by interpretation of these events in their context. FIDIS deliverable 7.2 has previously examined the data mining techniques which could be adopted for the data processing phase. 

From a purely pragmatic viewpoint, the agent needs to interact with or ‘read’ from the environment (for example from the thermostatic value in the example given in section ) to retrieve data and with services to provide the user with the desired level of support. In some cases, such as ‘online profiling’ it is possible to easily compile personal data from Internet Protocol traffic (e.g. domain names visited, internet services used, etc.), from website log files (e.g. time, history of visited pages or images downloaded), from internet cookies (useful information in order to recognise the visitor, e.g. registration information, number of customer, tracking of activities, etc.) or simply from user data input to websites. The relative ease of this data collection is a result of the standardised methodologies that are employed for interacting online. However, data collection within the AmI environment becomes significantly more complex because it is the complex interaction with a large variety of objects and services within a specific context that requires analysis. In the most part, these objects and services currently are not networked in any way at all.

As such, within the AmI environment, it is clear that a seamless interoperable data flow between the various interaction levels shown in , by proprietary devices from varying manufacturers, needs to be realised. This is only possible if the integrating network technology used is able to support systems integration, i.e. through standardised protocols. Some standards already exist such as CC/PP (Composite Capabilities / Preferences Profile) from the World Wide Web Consortium (W3C) and UAProf (User Agent Profile) proposed by the Wireless Access Protocol (WAP) forum. Notably, CC/PP is a standard based on Resource Description Framework (RDF), and is already used. Additionally, a number of standards for open communication in sensor networks have been proposed. Efforts to make buildings smarter are focusing on cutting costs by streamlining building operations, such as lighting and air conditioning. The most common networks are the BACnet and LonWorks standards, developed for building automation. These standards are geared towards well-defined application areas, and are built on top of well defined network structures. The net result of being so application specific is that many of the visions for AmI cannot be readily implemented on such systems. In practice, from the technical viewpoint, this may indicate that the AmI scenario is already running into issues of interoperability. A discussion of the sensor technologies that may be employed within the AmI network structure will be given in future FIDIS deliverables, starting with FIDIS deliverable 7.7 on RFID, AmI and profiling.

From a technical perspective, the process of forming a profile - even given access to the information required from the environment - is a computationally expensive process. Although developments in computing power are clear, it is supposed that the data mining of this vast amount of information will actually be implemented in distributed computing structures. As such, a standardised and interoperable agent that is abstracted from the underlying hardware such that it operates irrespective of the medium it is running on is required.  

It is reasonable to note here that ultimately the price for such increased data flow is likely to be borne by a decrease in potential user privacy as personal information becomes readily and widely linkable. In chapter some of the questions regarding privacy and security of this data are surveyed, while in chapter some proposals are made to store the resulting profiles on the user’s own personal digital assistant.