Structured Relation Discovery using Generative Models

Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1456–1466, Edinburgh, Scotland, UK, July 27–31, 2011. c⃝2011 Association for Computational Linguistics Structured Relation Discovery using Generative Models Limin Yao∗ Aria Haghighi+ Sebastian Riedel∗ Andrew McCallum∗ ∗ Department of Computer Science, University of Massachusetts at Amherst + CSAIL, Massachusetts Institute of Technology {lmyao,riedel,mccallum}@cs.umass.edu {aria42}@csail.mit.edu Abstract We explore unsupervised approaches to rela- tion extraction between two named entities; for instance, the semantic bornIn relation be- tween a person and location entity. Con- cretely, we propose a series of generative probabilistic models, broadly similar to topic models, each which generates a corpus of ob- served triples of entity mention pairs and the surface syntactic dependency path between them. The output of each model is a cluster- ing of observed relation tuples and their as- sociated textual expressions to underlying se- mantic relation types. Our proposed models exploit entity type constraints within a relation as well as features on the dependency path be- tween entity mentions. We examine effective- ness of our approach via multiple evaluations and demonstrate 12% error reduction in preci- sion over a state-of-the-art weakly supervised baseline. 1 Introduction Many NLP applications would benefit from large knowledge bases of relational information about entities. For instance, knowing that the entity Steve Balmer bears the leaderOf relation to the entity Microsoft, would facilitate question answer- ing (Ravichandran and Hovy, 2002), data mining, and a host of other end-user applications. Due to these many potential applications, relation extrac- tion has gained much attention in information ex- traction (Kambhatla, 2004; Culotta and Sorensen, 2004; Mintz et al., 2009; Riedel et al., 2010; Yao et al., 2010). We propose a series of generative prob- abilistic models, broadly similar to standard topic models, which generate a corpus of observed triples of entity mention pairs and the surface syntactic de- pendency path between them. Our proposed mod- els exploit entity type constraints within a relation as well as features on the dependency path between entity mentions. The output of our approach is a clustering over observed relation paths (e.g. “X was born in Y” and “X is from Y”) such that expressions in the same cluster bear the same semantic relation type between entities. Past work has shown that standard supervised techniques can yield high-performance relation de- tection when abundant labeled data exists for a fixed inventory of individual relation types (e.g. leaderOf) (Kambhatla, 2004; Culotta and Sorensen, 2004; Roth and tau Yih, 2002). However, less ex- plored are open-domain approaches where the set of possible relation types are not fixed and little to no labeled is given for each relation type (Banko et al., 2007; Banko and Etzioni, 2008). A more re- lated line of research has explored inducing rela- tion types via clustering. For example, DIRT (Lin and Pantel, 2001) aims to discover different repre- sentations of the same semantic relation using dis- tributional similarity of dependency paths. Poon and Domingos (2008) present an Unsupervised se- mantic parsing (USP) approach to partition …

Content truncated to 500 words