Optimizations for Answering Conjunctive ABox Queries

Conjunctive query answering is an important task for many applications on Semantic Web. It is important to efficiently answer queries over knowledge bases with large ABoxes. Although answering conjunctive queries has been studied from a theoretical point of view, until very recently, there was no reasoner supporting such functionality. As a result, there is not enough implementation experience and optimization techniques developed far answering conjunctive queries. In this paper, we address the problem of answering conjunctive ABox queries efficiently without compromising soundness and completeness. Our focus is on answering queries asked against a very large ABox. We consider queries with only distinguished variables. We focus on the case of distinguished variables because such queries can be answered more efficiently and such queries occur more frequently in realistic scenarios. We start with the discussion of answering atomic ABox queries and then describe a sound and complete conjunctive query answering algorithm. We then present optimization methods to improve the query evaluation analogous to the optimization methods developed and studied in the context of relational databases. We discuss how the ordering of query evaluation can affect performance, describe some simple yet effective metrics of evaluating the cost of a given ordering and present some heuristics to find (near-)optimal orderings. In the end, we show that the cost model we described provides a sufficiently accurate approximation to the actual query answering time.