EVALUATING THE COREF-PRO CROSS-DOCUMENT COREFERENCE SYSTEM
Nghiem Tri Duc, Bernardo Magnini, Roberto Zanoli · 2009
Cross-document coreference (CDC) plays an important role in text processing. It allows users to extract information about a particular entity from different documents. I evaluated a system for cross-document coreference developed in FBK that, differently for other systems needing a fixed threshold to group names referring to the same entity, tries first to guess the correct number of entities by evaluating the clusters quality, and then to co-refer person names to these entities. The evaluation phase included testing the system on an Italian dataset (Live Memories corpus) and an English one (Web People Search corpus).