Consolidating Entity Information from Heterogeneous Text Sources for Multi-INT Fusion

In this project we propose to develop technology for computing the similarity between two Entity Profiles compiled from text sources by an Information Extraction (IE) engine. The similarity computation is based on contextual information from the text and also on the full range of IE results (entity attributes, relationships and events). We identify several types of features across the spectrum between readily extracted but imprecise information and more precise but scarcer information types. A machine learning based classifier is used to integrate similarity measures across these information types into a single output that determines whether the two profiles are about the same entity and should be merged. The resulting similarity measures can be used to implement Name Disambiguation for entity profile merging and also user-initiated searches for entities most similar to a given exemplar.