Shifting Models Left: End-to-End Traceability as a Foundation for Knowledge Graphs as Data Products

Lionel Tailhardat, Joanna Balcerzack, Arij Elmajed, Pano Maria, Gabriel FOMI NGAMENI

Keywords: Semantic integration, Data lineage, Schema matching, Data mesh, RML, SPARQL, Formal proof

TL;DR: We introduce the end-to-end traceability hypothesis as a principled basis for elevating KGC pipelines from craftsmanship-driven integration efforts to auditable data products.

Abstract:

Various methods and tools already exist in the field of knowledge graph construction, which can be combined into pipelines for knowledge graph-based decision support systems. However, the semantic integration of multiple data sources into a knowledge graph is a craftsmanship task closely linked to the knowledge/data engineers' deep understanding of the information system they work in.
This vision paper proposes to approach the problem from the opposite side of the pipeline, rather than assembling integration logic bottom-up, we start from the hypothesis of end-to-end traceability, the verification mechanism that makes ``shifting semantics left'' operational, and views the knowledge graph as a data product. Our proposals are grounded in the logical nature of Semantic Web technologies and the concept of formal proof of systems.

Submission Number: 1