Open Relation Extraction by Matrix Factorization and Universal Schemas

Abstract

Universal schemas are a remarkable approach used to solve relation extraction, constructed by a union of all usable schemas. However, in free word order languages where surface form predicates are difficult to extract, original universal schemas cannot be applied. We introduce a novel extension: using dependency paths and entity types instead of surface form predicates. Based on the extension, we could cluster similar ontological relations, which broaden the coverage of question answering. We verified the performance of our extended model by constructing and evaluating our model in Korean based on Korean DBpedia and Korean Wikipedia distant supervision data.