Adding High-Precision Links to Wikipedia

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 651–656, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Adding High-Precision Links to Wikipedia Thanapon Noraset Chandra Bhagavatula Doug Downey Department of Electrical Engineering & Computer Science Northwestern University Evanston, IL 60208 {nor|csbhagav}@u.northwestern.edu, ddowney@eecs.northwestern.edu Abstract Wikipedia’s link structure is a valuable resource for natural language processing tasks, but only a fraction of the concepts mentioned in each article are annotated with hyperlinks. In this paper, we study how to augment Wikipedia with additional high-precision links. We present 3W, a system that identifies concept mentions in Wikipedia text, and links each mention to its referent page. 3W leverages rich semantic information present Wikipedia to achieve high precision. Ourin experiments demonstrate that 3W can add an average of seven new links to each Wikipedia article, at a precision of 0.98. 1 Introduction Wikipedia forms a valuable resource for many Natural Language Processing and Information Extraction tasks, such as Entity Linking (Cucerzan, 2007; Han and Zhao, 2009), Ontology Construction (Wu and Weld, 2008; Syed et al., 2008) and Knowledge Base Population (Hoffart et al., 2013; Lehmann et al., 2013). Wikipedia’s links provide disambiguated semantic information. For example, when a system processes the text “Chicago was received with critical acclaim” from an article, the system does not need to infer the referent entity of “Chicago” if the word is already hyperlinked to the Wikipedia page of the Oscar-winning film. Unfortunately, in Wikipedia only a fraction of the phrases that can be linked are in fact annotated with a hyperlink. This is due to Wikipedia’s conventions of only linking to each concept once, and only when the links have a certain level of utility for human readers.1 We see this as an 1http://en.wikipedia.org/wiki/ Wikipedia:Manual_of_Style_(linking) opportunity to improve Wikipedia as a resource for NLP systems. Our experiments estimate that as of September 2013, there were an average of 30 references to Wikipedia concepts left unlinked within each of English Wikipedia’s four million pages. In this paper, our goal is to augment Wikipedia with additional high-precision links, in order to provide a new resource for systems that use Wikipedia’s link structure as a foundation. Identifying references to concepts (called mentions) in text and linking them to Wikipedia is a task known as Wikification. Wikification for general text has been addressed in a wide variety of recent work (Mihalcea and Csomai, 2007; Milne and Witten, 2008b; McNamee and Dang, 2009; Ratinov et al., 2011). The major challenge of this task is to resolve the ambiguity of phrases, and recent work makes use of various kinds of information found in the document to tackle the challenge. In contrast to this body of work, here we focus on the special case of Wikifying Wikipedia articles, instead of general documents. This gives us an advantage over general-text systems due to Wikipedia’s rich content and existing link structure. We introduce 3W, a system that identifies mentions within Wikipedia and links each to its referent concept. We …

Content truncated to 500 words