Answering top-k queries using views

Abstract

The problem of obtaining efficient answers to top-k queries has attracted a lot of research attention. Several algorithms and numerous variants of the top-k retrieval problem have been introduced in recent years. The general form of this problem requests the k highest ranked values from a relation, using monotone combining functions on (a subset of) its attributes.In this paper we explore space performance tradeoffs related to this problem. In particular we study the problem of answering top-k queries using views. A view in this context is a materialized version of a previously posed query, requesting a number of highest ranked values according to some monotone combining function defined on a subset of the attributes of a relation. Several problems of interest arise in the presence of such views. We start by presenting a new algorithm capable of combining the information from a number of views to answer ad hoc top-k queries. We then address the problem of identifying the most promising (in terms of performance) views to use for query answering in the presence of a collection of views. We formalize both problems and present efficient algorithms for their solution. We also discuss several extensions of the basic problems in this setting.We present the results of a thorough experimental study that deploys our techniques on real and synthetic data sets. Our results indicate that the techniques proposed herein comprise a robust solution to the problem of top-k query answering using views, gracefully exploring the space versus performance tradeoffs in the context of top-k query answering.