Mining Information for Instance Unification

Instance unification determines whether two instances in an ontology refer to the same object in the real world. More specifically, this paper addresses the instance unification problem for person names. The approach combines the use of citation information (i.e., abstract, initials, titles and co-authorship information) with web mining, in order to gather additional evidence for the instance unification algorithm. The method is evaluated on two datasets – one from the BT digital library and one used in previous work on name disambiguation. The results show that the information mined from the web contributes substantially towards the successful handling of highly ambiguous cases which lowered the performance of previous methods.