Dynamic Parameters for Cross Document Coreference

In this paper we present a new algorithm for the Person Cross Document Coreference task. We show that accurate results require a way to adapt the parameters of the similarity function – metrics and threshold – to the ontological constraints obeyed by individuals. The technique we propose dynamically changes the initial weights computed when the context is analyzed. The weight recomputation is necessary in order to resolve clusters borders, which are inevitably blurred by a static approach. The results show a significant gain in accuracy.