A method of segmenting relative user preferences into fine-grain and coarse-grain collections is provided. The method includes providing a set of content items having associated descriptive terms. The method also includes receiving user search input and, in response thereto, presenting a subset of content items. The method includes receiving user selection actions and analyzing the selections to learn the user's preferred descriptive terms. The method includes expressing the learned preferred descriptive terms as a segmented probability distribution function having at least one fine grain segment and at least one coarse grain segment. In response to subsequent search input, the method calls for selecting and ordering a collection of content items by promoting the ranking of content items associated with the learned preferred descriptive terms of the user according to the differentiation provided by the segmented probability distribution function.