| Abstract |
Knowledge graphs capture complex domain-relevant information and relationships between different entities, often based on detailed ontologies. Knowledge graph embedding (KGE) can be used to encode data in a semantic vector space that captures the relative meaning of these elements based on their actual usage in a database. KGE techniques leverage latent information arising from topological similarities of sub-structures within the graph to deepen insight gained in downstream analytics tasks. To maximise the likely contextual relevance of the results, new techniques are being developed that take advantage of human-defined ‘meta-paths’ that enable context-sensitive random-walks through heterogeneous graph databases, corresponding to identified routes that make sense in the context of the analytics. To efficiently leverage these meta-paths from human users, human-agent knowledge fusion (HAFK) techniques can be quickly re-purposed to support sensemaking. We illustrate the potential for an integrated platform that rapidly enables user-machine information fusion by enabling downstream analytics and visualization to inform user input in a feedback loop. |