DC Proposal: Ontology Learning from Noisy Linked Data

Ontology learning - loosely, the process of knowledge extraction from diverse data sources - provides (semi-) automatic support for ontology construction. As the 'Web of Linked Data' vision of the Semantic Web is coming true, the 'explosion' of Linked Data provides more than sufficient data for ontology learning algorithms in terms of quantity. However, with respect to quality, notable issue of noises (e.g., partial or erroneous data) arises from Linked Data construction. Our doctoral researches will make theoretical and engineering contribution to ontology learning approaches for noisy Linked Data. More exactly, we will use the approach of Statistical Relational Learning (SRL) to develop learning algorithms for the underlying tasks. In particular, we will learn OWL axioms inductively from Linked Data under probabilistic setting, and analyze the noises in the Linked Data on the basis of the learned axioms. Finally, we will make the evaluation on proposed approaches with various experiments.