UMass CIIR at TAC KBP 2013 Entity Linking: Query Expansion using Urban Dictionary

This paper describes the system submitted to the TAC 2013 entity linking task of the Knowledge Base Population track. The core of the approach is probabilistic information retrieval over a search index of the knowledge base, including the text of Wikipedia. The retrieval results are further reranked using a supervised learning-to-rank model. The submission this year builds on the neighborhood and context extraction methods for query expansion introduced in 2012. In 2013, two new models are added. The first is query expansion using Urban Dictionary. For mentions in forum documents, we search Urban Dictionary for the mention string and perform query expansion. The second method is a multi-pass linking model. Instead of linking only the query mention, all entity mentions in the document are linked with features that encourage coherence among the linked entities. The results show that the method incorporating Urban Dictionary expansion run performed the best.