Unsupervised Relation Extraction with General Domain Knowledge

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 415–425, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Unsupervised Relation Extraction with General Domain Knowledge Oier Lopez de Lacalle1,2 and Mirella Lapata1 1Institute for Language, Cognition and Computation School of Informatics, University of Edinburgh, 10 Crichton Street, Edinburgh EH8 9AB 2IKERBASQUE, Basque Foundation for Science, Bilbao, Spain oier.lopezdelacalle@ehu.es, mlap@inf.ed.ac.uk Abstract In this paper we present an unsupervised ap- proach to relational information extraction. Our model partitions tuples representing an observed syntactic relationship between two named entities (e.g., “X was born in Y” and “X is from Y”) into clusters correspond- ing to underlying semantic relation types (e.g., BornIn, Located). Our approach incor- porates general domain knowledge which we encode as First Order Logic rules and auto- matically combine with a topic model devel- oped specifically for the relation extraction task. Evaluation results on the ACE 2007 English Relation Detection and Categoriza- tion (RDC) task show that our model outper- forms competitive unsupervised approaches by a wide margin and is able to produce clus- ters shaped by both the data and the rules. 1 Introduction Information extraction (IE) is becoming increas- ingly useful as a form of shallow semantic analy- sis. Learning relational facts from text is one of the core tasks of IE and has applications in a variety of fields including summarization, question answer- ing, and information retrieval. Previous work (Sur- deanu and Ciaramita, 2007; Culotta and Sorensen, 2004; Zhou et al., 2007) has traditionally relied on extensive human involvement (e.g., hand-annotated training instances, manual pattern extraction rules, hand-picked seeds). Standard supervised techniques can yield high performance when large amounts of hand-labeled data are available for a fixed in- ventory of relation types (e.g., Employment, Lo- cated), however, extraction systems do not easily generalize beyond their training domains and often must be re-engineered for each application. Un- supervised approaches offer a promising alternative which could lead to significant resource savings and more portable extraction systems. It therefore comes as no surprise that latent topic analysis methods have been used for a variety of IE tasks. Yao et al. (2011), for example, propose a series of topic models which perform relation discovery by clustering tuples representing an ob- served syntactic relationship between two named en- tities (e.g., “X was born in Y” and “X is from Y”). The clusters correspond to semantic relations whose number or type is not known in advance. Their mod- els depart from standard Latent Dirichlet Allocation (Blei et al., 2003) in that a document consists of re- lation tuples rather than individual words; moreover, tuples have features each of which is generated in- dependently from a hidden relation (e.g., the words corresponding to the first and second entities, the type and order of the named entities). Since these features are local, they cannot capture more global constraints pertaining to the relation extraction task. Such constraints may take the form of restrictions on which tuples should be clustered together …

Content truncated to 500 words