Named Entity Disambiguation at Scale

Abstract

Named Entity Disambiguation (NED) is a crucial task in many Natural Language Processing applications such as entity linking, record linkage, knowledge base construction, or relation extraction, to name a few. The task in NED is to map textual variations of a named entity to its formal name. It has been shown that parameter-less models for NED do not generalize to other domains very well. On the other hand, parametric learning models do not scale well when the number of formal names expands above the order of thousands or more. To tackle this problem, we propose a deep architecture with superior performance on NED and introduce a strategy to scale it to hundreds of thousands of formal names. Our experiments on several datasets for alias detection demonstrate that our system is capable of obtaining superior results with a large margin compared to other state-of-the-art systems.

Extraction of Knowledge Models from Textbooks

“…Proprietary digital textbooks can usually be accessed through the libraries of educational institutions. Sites like Project Gutenberg 4 and Internet Archive Books 5 provide free textbooks legally. In addition, there is a recent trend to produce high-quality, peer-reviewed open textbooks.…”

“…Fang et al [93] x x x Kruit et al [159] x x Gao et al [110] x x x Gao et al [108] x x x x x Ramanathan et al [226] x x x Wu et al [293] x x x Tkaczyk et al [265] x x x x x Wu et al [290] x x x x x Bast & Korzen [26] x x x x Tuarob et al [269,270] x x x x Larrañaga et al [165] x x Wali et al [280] x Lopes et al [171,172] x Dwarakanath et al [82] x Wang et al [284,285] x x Mihalcea & Csomai [195] x x Medelyan et al [189] x x Milne & Witten [197] x x Mendes et al [192] / Daiber et al [63] x x Moro et al [204] x x Zhu & Iglesias [302] x x Aghaebrahimian & Cielieback [5] x x…”

“…Extraction of terms (Larrañaga et al [165]) and domain concepts (Xu et al [294]) attracts attention since the early 2000s. Linking of Named Entities were studied as early as 2007 (Mihalcea & Csomai [195]), and continue to be a topic of interest nowadays (Aghaebrahimian & Cieliebak [5]). Extraction of concept relations are more studied since 2016 (Chaplot et al [53] and Labutov et al [161]).…”

“…Its latest version, YAGO3, is available for download. KBpedia 5 is another project combining knowledge from several public knowledge bases: Wikipedia, Wikidata, schema.org, DBpedia, GeoNames, OpenCyc 6 , and UMBEL 7 . It includes 55 thousand reference concepts, links to about 32 million entities, and 5 thousand relations and properties.…”

“…At the same time, even a surface analysis of a typical textbook index can show that not all index entries are equally representative of the domain of the textbook 5 . From an epistemological perspective, an index of any document reflects not only the Textbooks expertise and efforts of its creator, but also the needs of a group of users for whom the index is created, and the task that these users are engaged in [129].…”