Filtering Inaccurate Entity Co-references on the Linked Open Data

The Linked Open Data LOD initiative relies heavily on the interconnections between different open RDF datasets where RDF links are used to connect resources. There has already been substantial research on identifying identity links between resources from different datasets, a process that is often referred to as co-reference resolution. These techniques often rely on probabilistic models or inference mechanisms to detect identity relations. However, recent studies have shown considerable inaccuracies in the LOD datasets that pertain to identity relations, e.g., owl:sameAs relations. In this paper, we propose a technique that evaluates existing identity links between LOD resources and identifies potentially erroneous links. Our work relies on the position and relevance of each resource with regards to the associated DBpedia categories modeled through two probabilistic category distribution and selection functions. Our experimental results show that our work is able to semantically distinguish inaccurate identity links even in cases when high syntactical similarity is observed between two resources.