Efficiently Evaluating Skyline Queries on RDF Databases

Abstract

Abstract. Skyline queries are a class of preference queries that compute the pareto-optimal tuples from a set of tuples and are valuable for multi-criteria decision making scenarios. While this problem has received significant attention in the context of single relational table, skyline queries over joins of multiple tables that are typical of storage models for RDF data has received much less attention. A naïve approach such as a join-first-skyline-later strategy splits the join and skyline computation phases which limit opportunities for optimization. Other existing techniques for multi-relational skyline queries assume storage and indexing techniques that are not typically used with RDF which would require a preprocessing step for data transformation. In this paper, we present an approach for optimizing skyline queries over RDF data stored using a vertically partitioned schema model. It is based on the concept of a "Header Point" which maintains a concise summary of the already visited regions of the data space. This summary allows some fraction of non-skyline tuples to be pruned from advancing to the skyline processing phase, thus reducing the overall cost of expensive dominance checks required in the skyline phase. We further present more aggressive pruning rules that result in the computation of near-complete skylines in significantly less time than the complete algorithm. A comprehensive performance evaluation of different algorithms is presented using datasets with different types of data distributions generated by a benchmark data generator.

Evaluating skyline queries over vertically partitioned tables

“…Although Balke et al's algorithms [1] evaluate Skyline queries over Web data sources, these algorithms may be adapted to return the Skyline over VPTs. On the other hand, Chen et al [4] proposed three algorithms which prune the search space by means of a headerpoint; a headerpoint is a virtual object that is built using the worst seen values in each VPT. Afterward, the algorithms in [1,4] discard dominated objects among the accessed ones.…”

“…On the other hand, Chen et al [4] proposed three algorithms which prune the search space by means of a headerpoint; a headerpoint is a virtual object that is built using the worst seen values in each VPT. Afterward, the algorithms in [1,4] discard dominated objects among the accessed ones. In general, the algorithms defined in [1,4] try to avoid the whole scanning of the VPTs.…”

“…Afterward, the algorithms in [1,4] discard dominated objects among the accessed ones. In general, the algorithms defined in [1,4] try to avoid the whole scanning of the VPTs.…”

“…In [1,4], the authors introduced algorithms to evaluate Skyline queries over VPTs. On one hand, Balke et al [1] introduced two algorithms which sequentially scans the VPTs in a round-robin fashion.…”

“…the state-of-art solutions [1,4]. We propose two algorithms based on a heuristics to calculate the Skyline set when data are distributed among several VPTs.…”