Stanford-UBC Entity Linking at TAC-KBP

Abstract

This paper describes the joint Stanford-UBC knowledge base population system for the entity linking task. We modified our 2009 approach, which was based on frequencies of Wikipedia back-links, providing a context-independent mapping from strings to possible Wikipedia titles. We then built on this foundation, taking into account named-entity recognition (NER) and coreference resolution information to disambiguate entities within a document. Simple heuristics — both context-independent and contextsensitive — were sufficient for our runs to score higher than the median entry. This year an additional challenge was organized, in which systems could not use the free text from the Wikipedia pages associated with the knowledge base nodes. Our contextindependent run outperforms all systems that participated, but unfortunately we did not submit to this subtrack.