Open domain knowledge extraction: inference on a web scale

Though ontologies are considered central to foster the Semantic Web effort, their practical application on a Web scale is limited by the difficulty of building and maintaining high-coverage ontologies, for any of the almost unbounded number of social and technical domains mirrored on the Web. In this paper we present Open Knowledge Extraction (Open KE), a novel paradigm that creates a bridge between information retrieval, taxonomy learning and automated reasoning. Open KE builds on recently published algorithms for Open Information Extraction (Open IE) and automated taxonomy learning, which were shown able to extract information on a Web scale basis in an unsupervised manner. The key idea of Open KE is to generalize Open IE's lexicalized extractions with an automatically learned taxonomy. Lexicalized extractions are transformed in logic predicates, and used to populate a Semantic Model. An inference engine is then used to perform inductions and deductions. In this paper we describe the knowledge extraction workflow in its entirety, and apply it to a large-scale experiment in the domains of Artificial Intelligence and Virology.