Approaching the Efficient Frontier: Cooperative Database Retrieval Using High-Dimensional Skylines

Cooperative database retrieval is a challenging problem: top k retrieval delivers manageable results only when a suitable compensation function (e.g. a weighted mean) is explicitly given. On the other hand skyline queries offer intuitive querying to users, but result set sizes grow exponentially and hence can easily exceed manageable levels. We show how to combine the advantages of skyline queries and top k retrieval in an interactive query processing scheme using user feedback on a manageable, representative sample of the skyline set to derive most adequate weightings for subsequent focused top k retrieval. Hence, each user’s information needs are conveniently and intuitively obtained, and only a limited set of best matching objects is returned. We will demonstrate our scheme’s efficient performance, manageable result sizes, and representativeness of the skyline. We will also show how to effectively estimate users’ compensation functions using their feedback. Our approach thus paves the way to intuitive and efficient cooperative retrieval with vague query predicates.