Knowledge Base Unification via Sense Embeddings and Disambiguation

Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 726–736, Lisbon, Portugal, 17-21 September 2015. c⃝2015 Association for Computational Linguistics. Knowledge Base Unification via Sense Embeddings and Disambiguation Claudio Delli Bovi Department of Computer Science Sapienza University of Rome dellibovi@di.uniroma1.it Luis Espinosa-Anke Department of Information and Communication Technologies Universitat Pompeu Fabra luis.espinosa@upf.edu Roberto Navigli Department of Computer Science Sapienza University of Rome navigli@di.uniroma1.it Abstract We present KB-UNIFY, a novel approach for integrating the output of different Open Information Extraction systems into a single unified and fully disambiguated knowledge repository. KB-UNIFY con- sists of three main steps: (1) disambigua- tion of relation argument pairs via a sense- based vector representation and a large unified sense inventory; (2) ranking of se- mantic relations according to their degree of specificity; (3) cross-resource relation alignment and merging based on the se- mantic similarity of domains and ranges. We tested KB-UNIFY on a set of four heterogeneous knowledge bases, obtain- ing high-quality results. We discuss and provide evaluations at each stage, and re- lease output and evaluation data for the use and scrutiny of the community1. 1 Introduction The breakthrough of the Open Information Ex- traction (OIE) paradigm opened up a research area where Web-scale unconstrained Information Extraction systems are developed to acquire and formalize large quantities of knowledge. How- ever, while successful, to date most state-of-the- art OIE systems have been developed with their own type inventories, and no portable ontologi- cal structure. In fact, OIE systems can be very different in nature. Early approaches (Etzioni et al., 2008; Wu and Weld, 2010; Fader et al., 2011) focused on extracting a large number of relations from massive unstructured corpora, mostly rely- ing on dependencies at the level of surface text. Systems like NELL (Carlson et al., 2010) com- bine a hand-crafted taxonomy of entities and re- lations with self-supervised large-scale extraction 1http://lcl.uniroma1.it/kb-unify from the Web, but they require additional process- ing for linking and integration (Dutta et al., 2014). More recent work has focused, instead, on deeper language understanding, especially at the level of syntax and semantics (Nakashole et al., 2012; Moro and Navigli, 2013). By leveraging semantic analysis, knowledge gathered from un- structured text can be adequately integrated and used to enrich existing knowledge bases, such as YAGO (Mahdisoltani et al., 2015), FREEBASE (Bollacker et al., 2008) and DBPEDIA (Lehmann et al., 2014). A large amount of reliable struc- tured knowledge is crucial for OIE approaches based on distant supervision (Mintz et al., 2009; Riedel et al., 2010), even when multi-instance multi-learning algorithms (Surdeanu et al., 2012) or matrix factorization techniques (Riedel et al., 2013; Fan et al., 2014) come into play to deal with noisy extractions. For this reason a recent trend of research has focused on Knowledge Base (KB) completion (Nickel et al., 2012; Bordes et al., 2013), exploiting the fact that distantly super- vised OIE and structured knowledge can comple- ment each other. However, the majority of integra- tion approaches nowadays are not designed to deal …

Content truncated to 500 words