Personalized Page Rank for Named Entity Disambiguation

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 238–243, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Personalized Page Rank for Named Entity Disambiguation Maria Pershina Yifan He Ralph Grishman Computer Science Department New York University New York, NY 10003, USA {pershina,yhe,grishman}@cs.nyu.edu Abstract The task of Named Entity Disambiguation is to map entity mentions in the document to their correct entries in some knowledge base. We present a novel graph-based dis- ambiguation approach based on Personalized PageRank (PPR) that combines local and global evidence for disambiguation and ef- fectively filters out noise introduced by in- correct candidates. Experiments show that our method outperforms state-of-the-art ap- proaches by achieving 91.7% in micro- and 89.9% in macroaccuracy on a dataset of 27.8K named entity mentions. 1 Introduction Name entity disambiguation (NED) is the task in which entity mentions in a document are mapped to real world entities. NED is both useful on its own, and serves as a valuable component in larger Knowl- edge Base Construction systems (Mayfield, 2014). Since the surge of large, publicly available knowl- edge bases (KB) such as Wikipedia, the most popu- lar approach has been linking text mentions to KB nodes (Bunescu and Pas¸ca, 2006). In this paradigm, the NED system links text mentions to the KB, and quite naturally utilizes information in the KB to support the linking process. Recent NED systems (Cucerzan, 2007; Ratinov et al., 2011; Alhelbawy and Gaizauskas, 2014) usually exploit two types of KB information: local information, which measures the similarity between the text mention and the a candidate KB node; and global information, which measures how well the candidate entities in a docu- ment are connected to each other, with the assump- tion that entities appearing in the same document should be coherent. Both types of features have their strengths and drawbacks: local features better en- code similarity between a candidate and a KB node, but overlook the coherence between entities; global features are able to exploit interlinking information between entities, but can be noisy if they are used by their own, without considering information from the text and the KB (cf. Section 4). In this paper, we propose to disambiguate NEs us- ing a Personalized PageRank (PPR)-based random walk algorithm. Given a document and a list of en- tity mentions within the document, we first construct a graph whose vertices are linking candidates and whose edges reflects links in Wikipedia. We run the PPR algorithm on this graph, with the constraint that we only allow the highest scored candidate for each entity to become the start point of a hop. As all can- didates but the correct one are erronous and probably misleading, limiting the random walk to start from the most promising candidates effectively filters out potential noise in the Personalized PageRank pro- cess. Our method has the following properties: 1) as our system is based on a random walk algorithm, it does not require training …

Content truncated to 500 words