Dissimilarity-based approach for Identity Link Invalidation

Abstract

More and more datasets are currently connected by identity links using properties such as owl:sameAs expressed in OWL. Identity links are statements that declare that two resources refer to the same real-world entity. However, we cannot attest the correctness of all identity links. Without a central name authority, most identity links are generated by heuristics and they are not reviewed by experts. The main issue in invalidating identity links is the heterogeneity of datasets, they commonly do not share the same predicates. Furthermore, the description of the resources can be incomplete. Despite how the resources are described, identity links are necessary to link data for posterior use. In this paper, we present a framework to invalidate identity links by dissimilarity and outlier detection in equivalence classes of identity links.