Worst-case Optimal Query Answering for Greedy Sets of Existential Rules and Their Subclasses

TL;DR

This paper introduces a new class of existential rules called greedy bounded-treewidth sets (gbts), providing a worst-case optimal algorithm for query answering and classifying the complexity of various rule subclasses.

Contribution

It defines gbts rules, offers a generic optimal algorithm for query entailment, and classifies the complexity of multiple gbts subclasses.

Findings

The algorithm is worst-case optimal for combined, data, and query complexity.

Classified complexity of several gbts subclasses, expanding understanding of guarded rule fragments.

Provides a comprehensive complexity landscape for existential rule classes.

Abstract

The need for an ontological layer on top of data, associated with advanced reasoning mechanisms able to exploit the semantics encoded in ontologies, has been acknowledged both in the database and knowledge representation communities. We focus in this paper on the ontological query answering problem, which consists of querying data while taking ontological knowledge into account. More specifically, we establish complexities of the conjunctive query entailment problem for classes of existential rules (also called tuple-generating dependencies, Datalog+/- rules, or forall-exists-rules. Our contribution is twofold. First, we introduce the class of greedy bounded-treewidth sets (gbts) of rules, which covers guarded rules, and their most well-known generalizations. We provide a generic algorithm for query entailment under gbts, which is worst-case optimal for combined complexity with or without bounded predicate arity, as well as for data complexity and query complexity. Secondly, we classify several gbts classes, whose complexity was unknown, with respect to combined complexity (with both unbounded and bounded predicate arity) and data complexity to obtain a comprehensive picture of the complexity of existential rule fragments that are based on diverse guardedness notions. Upper bounds are provided by showing that the proposed algorithm is optimal for all of them.