Online maintenance of evolving knowledge graphs with RDFS-based saturation and why-provenance support

Abstract

Enterprise RDF knowledge graphs are often built using extraction data pipelines that are fed by several heterogeneous sources (relational databases, CSV files or even unstructured textual data). As a direct consequence, the construction of these KGs undergoes a number of changes in the early stages of their life cycle, which are initiated by a human developer and therefore need to be done interactively and efficiently. Driven by such needs, in this paper, we present a solution for the incremental maintenance of KGs given user-prescribed changes. A key feature of the proposed solution is the support of provenance collection that can be used to assist the developer in the analysis and debugging of the KG. Specifically, we strive to compute and maintain the provenance of asserted and inferred facts in the knowledge graph incrementally (and thus efficiently). The evaluation exercises we have conducted show the effectiveness of our solution and highlight the parameters that impact performance.