Utilising structured information for the representation and elicitation of user preferences

We make use of preferences every day. When we shop for a book or choose a meal; when we select music to listen to or which vacation destination to travel to on our summer holiday. Preferences are the key element in reasoning about these decisions. Therefore, the modeling and handling of preferences is an important challenge for the development of automated processes that aim to assist in the making of these decisions. Such decision assistance processes are often referred to as personalisation processes. The objective of this thesis is to demonstrate the value of utilising structured information in order to assist with the personalisation process.In some domains it may be difficult for people to express their preferences in terms of the domain attributes. Searching for items in these terms may either require some knowledge about the domain or require some technical familiarity with the items in the domain; knowledge which people do not always possess. In this case, preferences are more easily communicated by reflecting some feedback over actual items, rather than over abstract notions. However, eliciting user preferences in terms of these abstract and compact notions makes the reasoning with the user preferences more efficient. It is the responsibility of a personalisation system to elicit these preferences on behalf of the user in order to successfully produce accurate recommendations. In this thesis, we develop techniques for the representation of formal preferences and the elicitation of a diverse set of those preferences, which can be embedded within the normal user activity in a domain. By doing so, we aim to construct a preference profile on behalf of a user through a simple and effective interaction without requiring for any expert knowledge. We argue that this can be achieved by utilising structured domain information in the form of domain ontologies. We commence by investigating various methods for utilising the structure of a domain ontology to model similarity rankings. We then show how these rankings can be used as a heuristic for enriching a partial specification of a user preference profile. Finally, we show how such structured information can assist in guiding the elicitation of user preferences, making the personalisation process more effective while catering for a positive user experience