Knowledge Graphs are a powerful and increasingly popular approach to represent complex knowledge and to support analysis and the discovery of new facts. Knowledge Graphs offer themselves for the integration of insights from structured databases, manually curated data and batches of facts derived from large-scale document analysis. However, creating and maintaining large Knowledge Graphs is a laborious and therefore costly undertaking. We describe joint work from Fraunhofer SCAI and Kairntech from a project on enriching a Knowledge Graph on a specific therapeutic area – Schizophrenia and Bipolar Disorders – with new knowledge drawn from large amounts of scientific literature. Recent advances in NLP/AI allow for the automation of the underlying tasks of entity and relation extraction from text. As a result of the analysis, isolated facts from different publications are placed in a comprehensive network such that new insights can be derived: Where one paper may report that entity A affects entity B, another one may reveal that entity B is linked to entity C, but the connection between A and C may so far not have been made explicit anywhere. Large-scale literature analysis as suggested here renders many of these connections accessible. We address the underlying scientific motivation and state of the art as well as the respective contribution of the two partners and planned next steps. The presentation may be relevant for data science and information management experts in pharma, biotech and health as well as in the publishing sector and related domains.
