Generating a Large-Scale Entity Linking Dictionary from Wikipedia Link Structure and Article Text

Generating a Large-Scale Entity Linking Dictionary from Wikipedia Link Structure and Article Text Ravindra Harige, Paul Buitelaar Insight Centre for Data Analytics �National University of Ireland, Galway firstname.lastname � @insight-centre.org Abstract Wikipedia has been increasingly used as a knowledge base for open-domain Named Entity Linking and Disambiguation. In this task, a dictionary with entity surface forms plays an important role in finding a set of candidate entities for the mentions in text. Existing dictionaries mostly rely on the Wikipedia link structure, like anchor texts, redirect links and disambiguation links. In this paper, we introduce a dictionary for Entity Linking that includes name variations extracted from Wikipedia article text, in addition to name variations derived from the Wikipedia link structure. With this approach, we show an increase in the coverage of entities and their mentions in the dictionary in comparison to other Wikipedia based dictionaries. Keywords: Entity Linking, Wikipedia, DBpedia, Name Variation extraction 1. Introduction Entity Linking is the task of linking mentions in text to entities in a knowledge-base (KB) like Wikipedia and as- sign it an unambiguous entity (sense) identifier. This task is challenging and comprises of two main steps: i) iden- tifying phrases to annotate and ii) finding and linking the right entity in the KB, considering its context. These two steps are called phrase detection and disambiguation. In this work, we mainly focus on one specific subtask of the disambiguation step, namely candidate generation. Typically, a pre-computed dictionary derived from the KB is used for candidate generation as we do here with Wikipedia as a KB. The quality and scope of this dictio- nary plays an important role in the disambiguation process as its performance is not only dependent on the actual al- gorithm used to disambiguate but also on the quality and expressiveness of the dictionary used to determine the en- tity candidates for the phrase mentions in text. We use patterns to analyse the leading sentences of Wikipedia articles and query structured infobox informa- tion captured by DBpedia to gather name variations for entities, extended it with name variations mined from the Wikipedia link structure. Additionally, we employ a method to consolidate name variations extracted from dif- ferent approaches into one set of mappings, which is essen- tially an entity linking dictionary. We provide a two-way mapping, such that an entity in the KB can be reached from an entity mention and all entity mentions can be queried for any given named entity in the KB. 2. Related Work Several approaches exist for generating an entity-linking dictionary from Wikipedia or other knowledge resources. We review some of the well known entity linking dictionar- ies in the following. Google released a cross-lingual dictionary for English Wikipedia concepts consisting of 175 million surface forms that refer to 7.6 million entities derived from Wikipedia article titles, anchor text from inter-Wikipedia links, non- Wikipedia web pages linking to Wikipedia articles and non- Wikipedia web pages to non-Wikipedia pages for topics that have corresponding Wikipedia articles(Spitkovsky and Chang, 2012). DBpedia …

Content truncated to 500 words