Robust Disambiguation of Named Entities in Text

Abstract

Disambiguating named entities in natural-language text maps mentions\nof ambiguous names onto canonical entities like people or places,\nregistered in a knowledge base such as DBpedia or YAGO. This paper\npresents a robust method for collective disambiguation, by\nharnessing context from knowledge bases and using a new form of\ncoherence graph. It unifies prior approaches into a comprehensive\nframework that combines three measures: the prior probability of an\nentity being mentioned, the similarity between the contexts of a\nmention and a candidate entity, as well as the coherence among\ncandidate entities for all mentions together. The method builds a\nweighted graph of mentions and candidate entities, and computes a\ndense subgraph that approximates the best joint mention-entity\nmapping. Experiments show that the new method significantly\noutperforms prior methods in terms of accuracy, with robust behavior\nacross a variety of inputs.