Abstract: Systems and methods are disclosed that facilitate categorization of preferences for items, and generation of recommendation based on such preferences. Specifically, users are enabled to identify elements of items, such as written works, that they prefer. Each element can generally refer to a specific aspect or portion of the written work, as opposed to describing the work as a whole. Users are further enabled to select descriptors for each element, which describe their preferences for the element. Thereafter, natural language recommendations can be generated from the selected elements and descriptors, and transmitted to additional other users or prospective users. The selected element and descriptor pairs may also be used to categorize the work, and to automatically generate recommendations for the work.