Entity Disambiguation Using a Markov-Logic Network

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 846–855, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Entity Disambiguation Using a Markov-Logic Network Hong-Jie Dai1,2 Richard Tzong-Han Tsai3* Wen-Lian Hsu1,2* 1Department of Computer Science, National Tsing Hua University, 300 No. 101, Section 2, Kuang-Fu Road, Hsinchu, Taiwan, R.O.C. 2Intelligent Agent Systems Lab., Institute of Information Science, Academia Sinica, 128 Academia Road, Sec.2, Nankang, Taipei, Taiwan, R.O.C. 3Department of Computer Science & Engineering, Yuan Ze University 135 Yuan-Tung Road, Chungli, Taoyuan, Taiwan, R.O.C. hongjie@iis.sinica.edu.tw thtsai@saturn.yzu.edu.tw hsu@iis.sinica.edu.tw Abstract* Entity linking (EL) is the task of linking a textual named entity mention to a knowledge base entry. It is a difficult task involving many challenges, but the most crucial problem is entity ambiguity. Traditional EL approaches usually employ different constraints and filtering techniques to improve performance. However, these constraints are executed in several different stages and cannot be used interactively. In this paper, we propose several a disambiguation formulae/features and employ Markov logic network to model interweaved constraints found in one type of EL, gene mention linking. To assess our systems effectiveness in different applications, we adopt two evaluation schemes: article-wide and instance-based precision/recall/F-measure. Experimental results show that our system outperforms the baseline systems and state-of-the-art systems under both evaluation schemes. 1 Introduction Entity linking (EL) is the task of linking a textual named entity mention to a knowledge base (KB) entry, such as linking a person/organization mention to its Wikipedia entry (Kulkarni, Singh et al. 2009; McNamee and Dang 2009). This task has broad applications and is important in information extraction (IE) and text mining. EL involves many challenges, but the most crucial problem in EL is entity ambiguity. Take the name John A. Smith for example. It might *Corresponding author appear in KB as John Alexander Smith, John Blair Smith, John D. Smith, etc. Entity disambiguation determines which of all the possible John Smith KB entries a given “John Smith” refers to. Several disambiguation approaches have been proposed to address the entity ambiguity problem. For example, Dredze et al. (2010) formulated the disambiguation task as a ranking problem and developed features to link entities to Wikipedia entries. Zhang et al. (2010) used an automatically generated corpus to train a binary classifier to reduce ambiguities. Dai et al. (2010) collected external knowledge for each entity and calculated likelihoods stating the similarity of the current text with the knowledge to improve the disambiguation performance. In addition to the entity ambiguity problem, the EL task in Text Analysis Conference (TAC) 2009 introduce the absence issue (McNamee and Dang 2009): for entities that have no corresponding entry in the KB a NIL should be returned. To deal with the absence issue, Bunescu and Pasca (2006) filtered out linked mentions whose scores are less than a fixed threshold. Li et al. (2009) trained a separate binary classifier to validate linked mentions. Dredze et al. (2010) treated the NIL as another KB entry candidate to train their EL ranking model. Most …

Content truncated to 500 words