Rule-based Reasoning on Massively Parallel Hardware

In order to enable the semantic web as well as other time critical semantic applications, scaleable reasoning mechanisms are indispensable. To address this issue, in this paper we propose a rule-based reasoning algorithm which explores the highly parallel hardware of modern processors. In contrast to other approaches of parallel reasoning, our algorithm works with rules that can be defined depending on the application scenario and thus is able to apply different semantics. Furthermore we show how vector-based operations can be used to implement a performant match algorithm. We evaluate our approach by applying the ρdf, RDFS and pD* rule sets to different data sets and compare our results with other recent work. The evaluation shows that our approach is up to 9 times faster depending on the rule set and the used ontology and is able for example to apply the ρdf rules to an ontology with 2.2 million triples (and 1.3 million inferred triples) in less than 6 seconds.