Passive learning of conditional preferences

This research is concerned with learning a model of a user’s behaviour to accurately make predictions that reflect the user’s decision making process. We focus on learning the user’s preferences or utility function in a passive setting and extend previous approaches by allowing for both trade-offs and conditional preferences. We investigate suitable preference representation methods, and frame this as an abductive logic programming approach which can make use of domainspecific background knowledge (e.g., expressed in first-order logic) to guide the learning process. Furthermore, we want to enhance this approach to a probabilistic framework by making use of recent advances in the area of Statistical Relation Learning in order to deal with the inherent uncertainties that result from making inferences from passive observations in real-world settings.