Fine-grained Semantic Typing of Emerging Entities

Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, pages 1488–1497, Sofia, Bulgaria, August 4-9 2013. c⃝2013 Association for Computational Linguistics Fine-grained Semantic Typing of Emerging Entities Ndapandula Nakashole, Tomasz Tylenda, Gerhard Weikum Max Planck Institute for Informatics Saarbr¨ucken, Germany {nnakasho,ttylenda,weikum}@mpi-inf.mpg.de Abstract Methods for information extraction (IE) and knowledge base (KB) construction have been intensively studied. However, a largely under-explored case is tapping into highly dynamic sources like news streams and social media, where new entities are continuously emerging. In this paper, we present a method for discovering and se- mantically typing newly emerging out-of- KB entities, thus improving the freshness and recall of ontology-based IE and im- proving the precision and semantic rigor of open IE. Our method is based on a prob- abilistic model that feeds weights into in- teger linear programs that leverage type signatures of relational phrases and type correlation or disjointness constraints. Our experimental evaluation, based on crowd- sourced user studies, show our method performing significantly better than prior work. 1 Introduction A large number of knowledge base (KB) con- struction projects have recently emerged. Promi- nent examples include Freebase (Bollacker 2008) which powers the Google Knowledge Graph, Con- ceptNet (Havasi 2007), YAGO (Suchanek 2007), and others. These KBs contain many millions of entities, organized in hundreds to hundred thou- sands of semantic classes, and hundred millions of relational facts between entities. However, de- spite these impressive advances, there are still ma- jor limitations regarding coverage and freshness. Most KB projects focus on entities that appear in Wikipedia (or other reference collections such as IMDB), and very few have tried to gather entities “in the long tail” beyond prominent sources. Vir- tually all projects miss out on newly emerging en- tities that appear only in the latest news or social media. For example, the Greenlandic singer Nive Nielsen has gained attention only recently and is not included in any KB (a former Wikipedia article was removed because it “does not indicate the im- portance or significance of the subject”), and the resignation of BBC director Entwistle is a recently new entity (of type event). Goal. Our goal in this paper is to discover emerg- ing entities of this kind on the fly as they become noteworthy in news and social-media streams. A similar theme is pursued in research on open infor- mation extraction (open IE) (Banko 2007; Fader 2011; Talukdar 2010; Venetis 2011; Wu 2012), which yields higher recall compared to ontology- style KB construction with canonicalized and se- mantically typed entities organized in prespecified classes. However, state-of-the-art open IE meth- ods extract all noun phrases that are likely to de- note entities. These phrases are not canonical- ized, so the same entity may appear under many different names, e.g., “Mr. Entwistle”, “George Entwistle”, “the BBC director”, “BBC head En- twistle”, and so on. This is a problem because names and titles are ambiguous, and this hampers precise search and concise results. Our aim is for all recognized and newly dis- covered …

Content truncated to 500 words