A lexico-semantic pattern language for learning ontology instances from text

The Semantic Web aims to extend the World Wide Web with a layer of semantic information, so that it is understandable not only by humans, but also by computers. At its core, the Semantic Web consists of ontologies that describe the meaning of concepts in a certain domain or across domains. The domain ontologies are mostly created and maintained by domain experts using manual, time-intensive processes. In this paper, we propose a rule-based method for learning ontology instances from text that helps domain experts with the ontology population process. In this method we define a lexico-semantic pattern language that, in addition to the lexical and syntactical information present in lexico-syntactic rules, also makes use of semantic information. We show that the lexico-semantic patterns are superior to lexico-syntactic patterns with respect to efficiency and effectivity. When applied to event relation recognition in text-based news items in the domains of finance and politics using Hermes, an ontology-driven news personalization service, our approach has a precision and recall of approximately 80% and 70%, respectively.