Module Extraction in Expressive Ontology Languages via Datalog Reasoning
Authors: Ana Armas Romero, Mark Kaminski, Bernardo Cuenca Grau, Ian Horrocks
JAIR 2016 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our evaluation on a wide range of ontologies confirms the feasibility and benefits of our approach in practice. |
| Researcher Affiliation | Academia | Ana Armas Romero EMAIL Mark Kaminski EMAIL Bernardo Cuenca Grau EMAIL Ian Horrocks EMAIL Department of Computer Science, University of Oxford, Wolfson Building, Parks Road, Oxford, OX1 3QD, UK |
| Pseudocode | No | The paper describes its methodology in detail in Sections 4, 5, 6, and 7, but does not present any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We have implemented a prototype system for module extraction in Java, called Pris M, which bundles RDFox as a black-box datalog reasoner (Motik et al., 2014). Pris M is available online under academic license.4 4 http://www.cs.ox.ac.uk/isg/tools/Pris M/ |
| Open Datasets | Yes | We have evaluated our system on a set of test ontologies identified in the work of Glimm, Horrocks, Motik, Stoilos, and Wang (2014) as non-trivial for reasoning. 5 The ontologies used in our experiments are available for download at https://krr-nas.cs.ox.ac.uk/2015/jair/Pris M/test Ontologies.zip. |
| Dataset Splits | Yes | Unlike the work of Del Vescovo et al. (2013), who defined random signatures simply as random subsets of the ontology signature, we extracted such signatures using a randomised graph sampling algorithm. We first represented the syntactic dependencies between symbols in the (normalised) ontology as a graph, and then traversed the graph in a randomised way until we visited a set number n of nodes. The symbols corresponding to the visited nodes were then taken as a random signature.7 The advantage of this approach is that it yields signatures that are semantically connected , which we believe is likely to be the case in practical applications. The number n was chosen by default as 0.1% of the total graph and then increased by up to two orders of magnitude in cases where the resulting signatures typically contained less than 15 predicates and thus were too small to provide additional information w.r.t. genuine signatures. |
| Hardware Specification | Yes | All experiments were performed on a server with 2 Intel Xeon E5-2670 2.60GHz processors, each of which has 8 physical cores that serve 2 virtual cores each, making a total of 32 virtual cores. In our experiments we allocated 90GB of RAM, and RDFox was always run on 16 threads. |
| Software Dependencies | No | We have implemented a prototype system for module extraction in Java, called Pris M, which bundles RDFox as a black-box datalog reasoner (Motik et al., 2014). |
| Experiment Setup | Yes | For each kind of signature and each ontology, we have considered a sample of 400 runs and averaged module sizes and module extraction times. ... RDFox was always run on 16 threads. |