Extracting Literal Assertions for DBpedia from Wikipedia Abstracts

Florian Schrage, Nicolas Heist, Heiko Paulheim

Abstract: Knowledge Graph completion deals with the addition of missing facts to knowledge graphs. While quite a few approaches exist for type and link prediction in knowledge graphs, the addition of literal values (also called instance or entity attributes) is not very well covered in the literature. In this paper, we present an approach for extracting numerical and date literal values from Wikipedia abstracts. We show that our approach can add 643k additional literal values to DBpedia at a precision of about 95%.