Detecting Identical Entities in the Semantic Web Data

Large amount of entities published by various sources inevitably introduces inaccuracies, mainly duplicated information. These can even be found within a single dataset. In this paper we propose a method for automatic discovery of identity relationship between two entities (also known as instance matching) in a dataset represented as a graph (e.g. in the Linked Data Cloud). Our method can be used for cleaning existing datasets from duplicates, validating of existing identity relationships between entities within a dataset, or for connecting different datasets using the owl:sameAs relationship. Our method is based on the analysis of sub-graphs formed by entities, their properties and existing relationships between them. It can learn a common similarity threshold for particular dataset, so it is adaptable to its different properties. We evaluated our method by conducting several experiments on data from the domains of public administration and digital libraries.