TAC Entity Linking by Performing Full-document Entity Extraction and Disambiguation.
Silviu Cucerzan · 2011
The paper describes the system submitted to TAC 2011 for the English entity linking task of the Knowledge Base Population track. Instead of focusing only on the provided target strings, this system extracts and disambiguates globally all entities from each target document and then maps the target string to one of the entities extracted from the document. The main features employed by the system are topics associated with the entities in the knowledge base, which are derived from Wikipedia categories, list pages, and lexicosyntactic link patterns. The submitted run achieved high scores on the official test data (accuracy of 86.8 % and B-cubed+ F-measure of 0.841). 1