Optimal Preference Elicitation for Skyline Queries over Categorical Domains

Abstract

Abstract. When issuing user-specific queries, users often have a vaguely defined information need. Skyline queries identify the most "interesting" objects for users' incomplete preferences, which provides users with intuitive query formulation mechanism. However, the applicability of this intuitive query paradigm suffers from a severe drawback. Incomplete preferences on domain values can often lead to impractical skyline result sizes. In particular, this challenge is more critical over categorical domains. This paper addresses this challenge by developing an iterative elicitation framework. While user preferences are collected at each iteration, the framework aims to both minimize user interaction and maximize skyline reduction. The framework allows to identify a reasonably small and focused skyline set, while keeping the query formulation still intuitive for users. All that is needed is answering a few well-chosen questions. We perform extensive experiments to validate the benefits of our strategy and prove that a few questions are enough to acquire a desired manageable skyline set.

Database preference queries—a possibilistic logic approach with symbolic priorities

“…See also [30] for further discussions. Lastly, all efforts made on this topic, focus only on numerical domains with inherent orders, and do not consider skyline queries over categorical domains (if we except the work done in [36]). …”

An efficient skyline framework for matchmaking applications

“…Techniques to evaluate skylines in subspaces of a given space (i.e., subsets of attributes) are proposed in Yuan et al (2005) and Pei et al (2005Pei et al ( , 2006, and they consider skylines in multiple subspaces simultaneously to identify better skyline items. In Lee et al (2008), the MaxPrune framework is proposed to minimize the number of preference elicitation steps for skyline computations over categorical attributes. Thus, owing to minimum preference elicitation from users the framework has a great advantage when the amount of users' preference is not adequate.…”

Personalizing queries based on networks of composite preferences

“…Acquiring user preferences is another great challenge. In the context of IR and Databases, preferences are either entered explicitly through a query interface or acquired at query time using a preference elicitation mechanism [Balke et al 2007;Lee et al 2008]. Long-term preferences can be learnt based on user feedback as well.…”