Knowledge graphs are the powerhorse of today's knowledge bases, from personal digital assistants such as Amazon Alexa and Apple Siri, over biomedical data for pharmaceutical research, to public linked data sets such as increasingly released by governments worldwide They represent information as a network formed by (subject, predicate, object) triples, where subjects and objects are entities and the predicate represents a relation between them They allow for complex querying, reasoning, and question answering

In recent years, methods for embedding knowledge graphs in metric spaces have been developed to further support their growing importance Knowledge graph embedding facilitates important tasks such as knowledge graph completion, entity resolution, and entity deduplication Even so, research around knowledge graph embedding is still in its infancy, and their accuracy remains limited This project will build on our state-of-the-art method Conditional Network Embedding (CNE) -- an embedding technique designed for standard networks instead of for the more powerful knowledge graphs The aim in this project is to take the insights gained in the development of CNE, and use them to develop a novel class of knowledge graph embedding methods with superior properties and accuracy on these important tasks