A Graph-Based Approach to Named Entity Categorization in Wikipedia Using Conditional Random Fields

Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pp. 649–657, Prague, June 2007. c⃝2007 Association for Computational Linguistics A Graph-based Approach to Named Entity Categorization in Wikipedia Using Conditional Random Fields Yotaro Watanabe, Masayuki Asahara and Yuji Matsumoto Graduate School of Information Science Nara Institute of Science and Technology 8916-5 Takayama, Ikoma, Nara, 630-0192, Japan {yotaro-w,masayu-a,matsu}@is.naist.jp Abstract This paper presents a method for catego- rizing named entities in Wikipedia. In Wikipedia, an anchor text is glossed in a linked HTML text. We formalize named en- tity categorization as a task of categorizing anchor texts with linked HTML texts which glosses a named entity. Using this repre- sentation, we introduce a graph structure in which anchor texts are regarded as nodes. In order to incorporate HTML structure on the graph, three types of cliques are defined based on the HTML tree structure. We pro- pose a method with Conditional Random Fields (CRFs) to categorize the nodes on the graph. Since the defined graph may in- clude cycles, the exact inference of CRFs is computationally expensive. We introduce an approximate inference method using Tree- based Reparameterization (TRP) to reduce computational cost. In experiments, our pro- posed model obtained significant improve- ments compare to baseline models that use Support Vector Machines. 1 Introduction Named and Numeric Entities (NEs) refer to proper nouns (e.g. PERSON, LOCATION and ORGANI- ZATION), time expressions, date expressions and so on. Since a large number of NEs exist in the world, unknown expressions appear frequently in texts, and they become hindrance to real-world text analysis. To cope with the problem, one effective ways to add a large number of NEs to gazetteers. In recent years, NE extraction has been performed with machine learning based methods. However, such methods cannot cover all of NEs in texts. Therefore, it is necessary to extract NEs from ex- isting resources and use them to identify more NEs. There are many useful resources on the Web. We fo- cus on Wikipedia1 as the resource for acquiring NEs. Wikipedia is a free multilingual online encyclope- dia and a rapidly growing resource. In Wikipedia, a large number of NEs are described in titles of ar- ticles with useful information such as HTML tree structures and categories. Each article links to other related articles. According to these characteristics, they could be an appropriate resource for extracting NEs. Since a specific entity or concept is glossed in a Wikipedia article, we can regard the NE extraction problem as a document classification problem of the Wikipedia article. In traditional approaches for doc- ument classification, in many cases, documents are classified independently. However, the Wikipedia articles are hypertexts and they have a rich structure that is useful for categorization. For example, hyper- linked mentions (we call them anchor texts) which are enumerated in a list tend to refer to the articles that describe other NEs belonging to the same class. It is expected that improved NE categorization is ac- complished …

Content truncated to 500 words