Discovering Relations between Noun Categories

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1447–1455, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Discovering Relations between Noun Categories Thahir P Mohamed * Estevam R Hruschka Jr. Tom M Mitchell University Of Pittsburgh Federal University of Sao Carlos Carnegie Mellon University pmthahir@gmail.com estevam@cs.cmu.edu tom.mitchell@cs.cmu.edu Abstract Traditional approaches to Relation Extraction from text require manually defining the rela- tions to be extracted. We propose here an ap- proach to automatically discovering relevant relations, given a large text corpus plus an ini- tial ontology defining hundreds of noun cate- gories (e.g., Athlete, Musician, Instrument). Our approach discovers frequently stated rela- tions between pairs of these categories, using a two step process. For each pair of categories (e.g., Musician and Instrument) it first co- clusters the text contexts that connect known instances of the two categories, generating a candidate relation for each resulting cluster. It then applies a trained classifier to determine which of these candidate relations is semanti- cally valid. Our experiments apply this to a text corpus containing approximately 200 million web pages and an ontology containing 122 cat- egories from the NELL system [Carlson et al., 2010b], producing a set of 781 proposed can- didate relations, approximately half of which are semantically valid. We conclude this is a useful approach to semi-automatic extension of the ontology for large-scale information extrac- tion systems such as NELL. 1 Introduction The Never-Ending Language Learner (NELL) (Carlson et al., 2010b)) is a computer system that learns continuously to extract facts from the web. NELL is given as input an initial ontology that specifies the semantic categories (e.g. city, compa- ny, sportsTeam) and semantic relations (e.g. hasOf- ficesIn(company,city), teamPlay- sInCity(sportsTeam,city)) it must extract from the web. In addition, it is provided 10-20 seed positive training examples for each of these categories and relations, along with hundreds of millions of unla- beled web page. Given this input, NELL applies a large-scale multitask, semisupervised learning method to learn to extract new instances of these categories (e.g., city(“London”)) and relations (e.g., teamPlaysInCity(“Steelers”,”Pittsburgh”)) from the web. During the past 17 months NELL has been running nearly continuously, learning to extract over 600 categories and relations, and pop- ulating a knowledge base containing over 700,000 instances of these categories and relations with a precision of approximately 0.851. This paper considers the problem of automati- cally discovering new relations to extend the on- tology of systems such as NELL, enabling them to increase over time their learning and extraction capabilities. More precisely, we consider the fol- lowing problem: Input: · An ontology specifying a set of categories · A knowledge base containing instances of these categories (perhaps including errors) · A large text corpus Output: · A set of two-argument relations that are fre- quently mentioned in the text corpus, and whose argument types correspond to categories in the input ontology (e.g., RiverFlows- ThroughCity(<River>,<City>). · For each proposed relation, a set of instances (i.e. RiverFlowsThroughCity(“Nile”,”Cairo”)). · For each proposed relation, …

Content truncated to 500 words