Augmenting Wikipedia with Named Entity Tags

Augmenting Wikipedia with Named Entity Tags Wisam Dakka Columbia University 1214 Amsterdam Avenue New York, NY 10027 wisam@cs.columbia.edu Silviu Cucerzan Microsoft Research 1 Microsoft Way Redmond, WA 98052 silviu@microsoft.com Abstract Wikipedia is the largest organized knowledge repository on the Web, increasingly employed by natural language processing and search tools. In this paper, we investigate the task of labeling Wikipedia pages with standard named entity tags, which can be used further by a range of in- formation extraction and language processing tools. To train the classifiers, we manually anno- tated a small set of Wikipedia pages and then ex- trapolated the annotations using the Wikipedia category information to a much larger training set. We employed several distinct features for each page: bag-of-words, page structure, ab- stract, titles, and entity mentions. We report high accuracies for several of the classifiers built. As a result of this work, a Web service that classi- fies any Wikipedia page has been made available to the academic community. 1 Introduction Wikipedia, one of the most frequently visited web sites nowadays, contains the largest amount of knowledge ever gathered in one place by volunteer contributors around the world (Poe, 2006). Each Wikipedia article contains information about one entity or concept, gathers information about entities of one particular type of entities (the so- called list pages), or provides information about homonyms (disambiguation pages). As of July 2007, Wikipedia contains close to two million articles in English. In addition to the English- language version, there are 200 versions in other languages. Wikipedia has about 5 million registered contributors, averaging more than 10 edits per contributor. Natural language processing and search tools can greatly benefit from Wikipedia by using it as an authoritative source of common knowledge and by exploiting its interlinked structure and disambiguation pages, or by extracting concept co- occurrence information. This paper presents a successful study on enriching the Wikipedia data with named entity tags. Such tags could be employed by disambiguation systems such as Bunescu and Pa�ca (2006) and Cucerzan (2007), in mining relationships between named entities, or in extracting useful facet terms from news articles (e.g., Dakka and Ipeirotis, 2008). In this work, we classify the Wikipedia pages into categories similar to those used in the CoNLL shared tasks (Tjong Kim Sang, 2002; Tjong Kim Sang and De Meulder, 2003) and ACE (Doddington et al., 2004). To the best of our knowledge, this is the first attempt to perform such classification on the English language version of the collection. 1 Although the task settings are different, the results we obtained are comparable with those previously reported in document classification tasks. We examined the Wikipedia pages to extract several feature groups for our classification task. We also observed that each entity/concept has at least two pseudo-independent views (page-based features and link-based features), which allow the use a co-training method to boost the performance of classifiers trained separately on each view. The classifier that achieved the best accuracy on out test set was applied then to all …

Content truncated to 500 words