Personalized Nonlinear Ranking Using Full-text Preferences

Today Internet systems commonly use a total ranking to present search results. These rankings are typically cut off at arbitrary points which are hard to understand. In this paper we present a new approach for rankings based on partial orders, which model personal preferences. It naturally groups large result sets according to the quality of results and presents only the top ones. It is possible for the user to expand these result sets selectively along chains of the partial order. We expect a considerable gain in comprehensibility, clarity and user friendliness. A pilot application is being implemented and first encouraging evaluation results are reported