A Graph-based Method for Entity Linking

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1010–1018, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP A Graph-based Method for Entity Linking Yuhang Guo, Wanxiang Che, Ting Liu∗, Sheng Li Research Center for Social Computing and Information Retrieval MOE-Microsoft Key Laboratory of Natural Language Processing and Speech School of Computer Science and Technology Harbin Institute of Technology, China {yhguo, car, tliu∗, sli}@ir.hit.edu.cn Abstract In this paper, we formalize the task of finding a knowledge base entry that a giv- en named entity mention refers to, name- ly entity linking, by identifying the most “important” node among the graph n- odes representing the candidate entries. With the aim of ranking these entities by their “importance”, we introduce three degree-based measures of graph connec- tivity. Experimental results on the TAC- KBP benchmark data sets show that our graph-based method performs comparably with the state-of-the-art methods. We al- so show that using the name phrase fea- ture outperforms the commonly used bag- of-word feature for entity linking. 1 Introduction Entity linking is the task of computationally map- ping a named entity mention in given context to the intended entry in a referential knowledge base. Large-scale knowledge bases have been proved to be valuable for many natural language pro- cessing applications such as question answering (MacKinnon and Vechtomova, 2008), informa- tion extraction (Pas¸ca, 2009), information retrieval (Santamar´ıa et al., 2010), coreference resolution (Ponzetto and Strube, 2007) and word sense dis- ambiguation (Fogarolli, 2009). And entity linking is a natural way to access these knowledge bases. Mention ambiguity is prevalent in language use. For example, in the following two sentences • “Mount Bromo is one of Java’s most popular tourist attractions.” ∗Corresponding author • “Candidates must have technical skills in JSP, ASP, Java, HTML.” the named entity mention Java refers to an island in Indonesia and a programming language respec- tively. In this paper, we focus on the named entity dis- ambiguation in entity linking and approach this problem from a graphical perspective. We begin by building a graph in which nodes correspond to the context around the target mention and the can- didate entries from the knowledge base, whereas directed edges represent the reference dependen- cy between nodes. Then we calculate the “im- portance” score for each candidate node and as- sign the most “important” candidate to the target mention. Here we compare three degree-based measures of graph connectivity that assess the n- ode importance. Through experiments performed on benchmark data sets, we show that the graph- based method achieves comparable performance to the state-of-the-art in the entity linking task. The result also indicates that to build linkage be- tween nodes, name phrase (i.e. n-gram of name words) performs better than the traditional bag- of-word feature. Our contributions are threefold: introducing an entity linking method under the graph-based framework, a novel in-degree graph connectivity measure for entity disambiguation, and an empirical comparison of the bag-of-word and the name phrase feature. This paper is organized as follows. Section …

Content truncated to 500 words