• View
• Edit
• History
• Print

A Neighborhood Relevance Model For Entity Linking

ABSTRACT

Entity Linking is the task of mapping mentions in documents to entities in a knowledge base. One of the crucial tasks is to identify the disambiguating context of the mention, and joint assignment models leverage the relationships within the knowledge base. Typical state-of-the-art approaches first heuristically identify candidates and then perform joint analysis. In contrast, we advocate to reason about the entity context already before the candidate generation phase. The goal is to obtain a small candidate set that contains the true answer with high likelihood, allowing for computationally intensive inference in the next phase. We achieve this by incorporating across-document evidence into an information retrieval model for candidates - retrieving and ranking candidate entities in one step. In combination with a discriminatively trained learning-to-rank approach our method is either outperforming or en pars with best performers of the TAC KBP competition in years 2009 - 2012.

BIO

Laura Dietz is a post-doctoral researcher working with Bruce Croft. Before that she was working with Andrew McCallum. She obtained her PhD on topic models for networked data from Max Planck Institute for Informatik in early 2011, being supervised by Tobias Scheffer and Gerhard Weikum.