A graphical method for reference reconciliation

Abstract

In many applications several references may refer to one real entity, the task of reference reconciliation is to group those references into several clusters so that each cluster associates with only one real entity. In this paper we propose a new method for reference reconciliation, that is, in addition to the traditional attribute values similarity, we employ the record-level relationships to compute the association similarity values of references in graphs, then we combine this kind of similarity with the traditional attribute values similarity and use the clustering algorithm to group the closest references. © 2010 Springer-Verlag Berlin Heidelberg.