Named Entity Disambiguation for little known referents: a topic-based approach

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1481–1492, Osaka, Japan, December 11-17 2016. Named Entity Disambiguation for little known referents: a topic-based approach Andrea Glaser and Jonas Kuhn Institute for Natural Language Processing University of Stuttgart, Germany andrea.glaser@ims.uni-stuttgart.de Abstract We propose an approach to Named Entity Disambiguation that avoids a problem of standard work on the task (likewise affecting fully supervised, weakly supervised, or distantly supervised machine learning techniques): the treatment of name mentions referring to people with no (or very little) coverage in the textual training data is systematically incorrect. We propose to in- directly take into account the property information for the “non-prominent” name bearers, such as nationality and profession (e.g., for a Canadian law professor named Michael Jackson, with no Wikipedia article, it is very hard to obtain reliable textual training data). The target property information for the entities is directly available from name authority files, or inferrable, e.g., from listings of sportspeople etc. Our proposed approach employs topic modeling to exploit tex- tual training data based on entities sharing the relevant properties. In experiments with a pilot implementation of the general approach, we show that the approach does indeed work well for name/referent pairs with limited textual coverage in the training data. 1 Introduction A central subtask for complex information retrieval and natural language understanding problems lies in the determination of what are the real-world entities that the proper names in a text refer to. While for some proper names (e.g., Henry VIII of England), the correct referent can be uniquely determined, the great majority of name mentions requires disambiguation. For instance, the name Michael Jackson can refer to the famous American singer, a British writer and beer expert, a Canadian actor, and many other people. The corresponding technical Natural Language Processing (NLP) Task, which is known by various names – Named Entity Disambiguation (e.g., Hoffart et al. (2011)), Entity Linking (e.g., Han et al. (2011)), or Wikification (Mihalcea and Csomai, 2007) – is typically construed as determining for each textual mention of a proper name, which of (typically) several entries in a knowledge base (such as DBpedia or Wikipedia) representing unique referents is the correct one in the given context. The standard approach to this task is to view it as a supervised classification problem, i.e., training data of textual mentions labeled with the correct disambiguation target are used to induce knowledge about indicative contextual clues for each candidate. A considerable amount of research has gone into the development of effective models (Mann and Yarowsky, 2003; Malin, 2005; Bollegala et al., 2006; Chen and Martin, 2007), and particularly into weakly supervised or distant supervision techniques, i.e., finding ways of exploiting explicit or implicit indications for the correct name references in real-life data (Cohen, 2005; Bunescu and Pasca, 2006; Cucerzan, 2007; Mihalcea and Csomai, 2007; Han et al., 2011; Hoffart et al., 2011). Indeed, for medium-to-high frequency name/referent combinations, it is not hard to harvest the web for suitable …

Content truncated to 500 words