Discovering and disambiguating named entities in text

Abstract

Disambiguating named entities in natural language texts maps ambiguous names to canonical entities registered in a knowledge base such as DBpedia, Freebase, or YAGO. Knowing the specific entity is an important asset for several other tasks, e.g. entity-based information retrieval or higher-level information extraction. Our approach to named entity disambiguation makes use of several ingredients: the prior probability of an entity being mentioned, the similarity between the context of the mention in the text and an entity, as well as the coherence among the entities. Extending this method, we present a novel and highly efficient measure to compute the semantic coherence between entities. This measure is especially powerful for long-tail entities or such entities that are not yet present in the knowledge base. Reliably identifying names in the input text that are not part of the knowledge base is the current focus of our work.