SAOR: Template Rule Optimisations for Distributed Reasoning over 1 Billion Linked Data Triples

Abstract

Abstract. In this paper, we discuss optimisations of rule-based materialisation approaches for reasoning over large static RDF datasets. We generalise and reformalise what we call the "partial-indexing" approach to scalable rule-based materialisation: the approach is based on a separation of terminological data, which has been shown in previous and related works to enable highly scalable and distributable reasoning for specific rulesets; in so doing, we provide some completeness propositions with respect to semi-naïve evaluation. We then show how related work on template rules -T-Box-specific dynamic rulesets created by binding the terminological patterns in the static ruleset -can be incorporated and optimised for the partial-indexing approach. We evaluate our methods using LUBM(10) for RDFS, pD* (OWL Horst) and OWL 2 RL, and thereafter demonstrate pragmatic distributed reasoning over 1.12 billion Linked Data statements for a subset of OWL 2 RL/RDF rules we argue to be suitable for Web reasoning.

Practical RDF Schema Reasoning with Annotated Semantic Web Data

“…The feasibility of large scale classical RDFS reasoning and its extensions has been shown in the literature [27,10,25,13,11]. In this paper we will show that the inclusion of annotated reasoning naturally introduces some overhead but that it is still possible to perform the closure of large annotated RDFS data in reasonable amount of time.…”

“…However, SOAR has more rules and implements a subset of OWL inferences, which we do not address here. It is used a technique call partial-indexing which relies on pre-processing a comparatively small-sized T-box with respect to the assertional data [11]. This terminological data extends the knowledge in our rdfs:subClassof, rdfs:subPropertyOf, rdfs:domain, and rdfs:range properties, and is not the major concern of the authors.…”

Large-Scale Storage and Reasoning for Semantic Data Using Swarms

“…Hogan et al [35] follow a pre-processing approach to scalable reasoning based on a semantics-preserving separation of terminological data. They create a set of stand-alone "template" rules formed from integrating the TBox into the reasoning rules.…”

Robust and scalable Linked Data reasoning incorporating provenance and trust annotations

“…In previous works, we presented our Scalable Authoritative OWL Reasoner (SAOR) which applies a subset of OWL 2 RL/RDF rules over arbitrary Linked Data crawls: in particular, we abandon completeness in favour of conducting "sensible" inferencing which we argue to be suitable for the Linked Data usecase [36,38]. 4 We have previously demonstrated distributed reasoning over ∼1b Linked Data triples [38], and have also presented some preliminary formalisations of what we call "authoritative reasoning", which considers the source of terminological data during reasoning [36].…”

“…4 We have previously demonstrated distributed reasoning over ∼1b Linked Data triples [38], and have also presented some preliminary formalisations of what we call "authoritative reasoning", which considers the source of terminological data during reasoning [36]. The SAOR system is actively used for materialising inferences in the Semantic Web Search Engine (SWSE) [37] which offers search and browsing over Linked Data.…”