Semantifying Triples from Open Information Extraction Systems

Arnab Dutta, Christian Meilicke, Heiner Stuckenschmidt

Abstract: The last few years have witnessed some remarkable success of the state-of-the art unsupervised knowledge extraction systems like NELL and REVERB. These systems are gifted with typically web-scale coverage but are often plagued with ambiguity due to lack of proper schema or unique identifiers for the instances. This classifies them apart from extraction systems like DBPEDIA, YAGO or FREE-BASE which have precise information content but have smaller coverage. In this work we bring together the former to enrich the later with high precision novel facts and present a statistical approach to discover new knowledge. In particular, we semantify NELL triples using DBPEDIA.