This website will look much better in a web browser that supports web standards, but it is accessible to any browser or Internet device. Go to main content.

ISWC2014 Tutorial on  Large Scale Reasoning Over Semantic Data

This page describes a full day tutorial at the 13th International Semantic Web Conference (ISWC 2014) (#ISWC2014) Riva del Garda, Trentino Italy, 19-20 Oct, 2014.


The tutorial aims to provide an overview of the approaches used for large scale reasoning over semantic data, the systems developed as well as the lessons learned while developing them. We will discuss some applications which require scalable reasoning solutions. Questions such as what makes distributed/parallel reasoning hard would also be covered during the tutorial. Directions for future research work would be discussed.

Software for download: TrOWL


1. Introduction (10m) - Jeff Z. Pan
Overview of the tutorial and introduction of speakers.

2. Overview of background knowledge (35m) - Jeff Z. Pan, Ilias Tachmazidis
We will introduce basics of linked data, (annotated) RDF(S), OWL and query answering. We will also give an introduction of MapReduce and other parallel frameworks.

3. Scalable OWL 2 DL reasoning based on approximation and divide-and-conquer approaches (45m) - Jeff Z. Pan
We will introduce two convincing approaches to support OWL reasoning. The idea of approximate reasoning is to simplify a given ontology to perform efficient reasoning for a given reasoning task. The idea of divide-and-conquer reasoning is to divide the computational load so as to improve the scalability and efficiency of semantic reasoning. We will also introduce how to ensure reasoning quality within these two approaches.

(Coffee break, 30m)

4. Scalable RDFS Reasoning Using MapReduce (45m) - Guilin Qi
Several extensions of RDF were proposed in order to deal with time, uncertainty, trust and provenance. All these specific domains can be modeled by a general framework called annotated RDF data. Along with discussing the MapReduce approach for scalable reasoning over annotated RDFS, we present some challenges that are unique to an arbitrary annotation domain. We will also show that our method is scalable for fuzzy logic, which is one of the annotated domains.

5. Distributed Reasoning in OWL 2 EL (45m) - Raghava Mutharaju
Existing reasoners make use of a single machine and are thus constrained by memory and computational power. Certain applications need reasoning support over streaming ontologies and also for knowledge bases which have been constructed automatically. In such cases, existing reasoners will be a limiting factor. In this part of the tutorial, we discuss approaches such as use of MapReduce framework and peer-to-peer systems for scalable reasoning over ontologies specific to OWL 2 EL profile.

(Lunch break, 60m)

6. Large Scale Non-Monotonic Reasoning (45m) - Ilias Tachmazidis
Monotonic reasoning does not perform well in the face of imperfect data, such as the ones available on the Web. Nonmonotonic reasoning is a prominent solution for processing missing or imprecise information. In particular, existing work on \emph{defeasible logic} and the \emph{well-founded semantics} have the ability to handle such datasets. In this part, we discuss the scalable reasoning approaches to such logics as well as the evaluation results that indicate the potential to reason over billions of facts.

7. Applications (45m) - Raghava Mutharaju, Jeff Z. Pan
There are some innovative and complex applications in the area of smart cities where temporal and spatial aspects have to be taken into consideration too. An even more challenging problem is to reason with dynamic (streaming) data. Current works on parallel reasoning based on MapReduce make sense when reasoning is done offline. When data changes, the results of previous reasoning cannot be reused and the systems have to do reasoning from scratch.

(Coffee break, 30m)

8. Lessons Learned (60m) - Jeff Z. Pan
Along with the success stories, there have been several not so successful attempts in scalable reasoning. We will discuss some of these attempts as well as the reasons which make distributed reasoning hard. Based on the attempts made so far at scalable reasoning, we will summarise the lessons learned. This will be very useful for researchers venturing into this topic.

Last modified 10:07:53 10/17/14