Inductive User Preference Manipulation for Multimedia Retrieval
David Zellhöfer · 2010
In order to enable users to query documents according their individual preferences, we propose a new user interaction model that forms an extension of the well-known relevance feedback approach. The introduced approach is utilizing partially ordered sets to express quality relations between result documents, i.e. the user's preference, directly on sample documents from the document set. Hence, the presented system supports users by offering an intuitive preference formulation which is known from daily life: the spontaneous quality judgement between objects without deeper knowledge of underlying attributes. This facilitates the interaction with the presented system as no new cognitive burdens are introduced into the search process. Based on these preferences, a machine-learning algorithm concludes an appropriate query via inductive reasoning in order to retrieve more relevant documents in an iterative manner. To conclude with, an initial prototype is discussed. First experiments show the utility of our approach.