Linking Entities to a Knowledge Base with Query Expansion

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 804–813, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Linking Entities to a Knowledge Base with Query Expansion Swapna Gottipati School of Information Systems Singapore Management University Singapore swapnag.2010@smu.edu.sg Jing Jiang School of Information Systems Singapore Management University Singapore jingjiang@smu.edu.sg Abstract In this paper we present a novel approach to entity linking based on a statistical lan- guage model-based information retrieval with query expansion. We use both local con- texts and global world knowledge to expand query language models. We place a strong emphasis on named entities in the local con- texts and explore a positional language model to weigh them differently based on their dis- tances to the query. Our experiments on the TAC-KBP 2010 data show that incor- porating such contextual information indeed aids in disambiguating the named entities and consistently improves the entity linking per- formance. Compared with the official re- sults from KBP 2010 participants, our system shows competitive performance. 1 Introduction When people read news articles, Web pages and other documents online, they may encounter named entities which they are not familiar with and there- fore would like to look them up in an encyclope- dia. It would be very useful if these entities could be automatically linked to their corresponding encyclo- pedic entries. This task of linking mentions of enti- ties within specific contexts to their corresponding entries in an existing knowledge base is called en- tity linking and has been proposed and studied in the Knowledge Base Population (KBP) track of the Text Analysis Conference (TAC) (McNamee and Dang, 2009). Besides improving an online surfer’s brows- ing experience, entity linking also has potential us- age in many other applications such as normalizing entity mentions for information extraction. The major challenge of entity linking is to resolve name ambiguities. There are generally two types of ambiguities: (1) Polysemy: This type of ambigu- ities refers to the case when more than one entity shares the same name. E.g. George Bush may re- fer to the 41st President of the U.S., the 43rd Presi- dent of the U.S., or any other individual who has the same name. Clearly polysemous names cause diffi- culties for entity linking. (2) Synonymy: This type of ambiguities refers to the case when more than one name variation refers to the same entity. E.g. Metro-Goldwyn-Mayer Inc. is often abbreviated as MGM. Synonymy affects entity linking when the en- tity mention in the document uses a name variation not covered in the entity’s knowledge base entry. Intuitively, to disambiguate a polysemous entity name, we should make use of the context in which the name occurs, and to address synonymy, exter- nal world knowledge is usually needed to expand acronyms or find other name variations. Indeed both strategies have been explored in existing litera- ture (Zhang et al., 2010; Dredze et al., 2010; Zheng et al., 2010). However, most existing work uses supervised learning approaches that require careful …

Content truncated to 500 words