CKG-P: A Prescriptive Process for Building Knowledge Graphs in a Compliance-Driven Corporate Environment

Abstract

Knowledge Graphs (KGs) have become increasingly critical in enterprise environments; however, existing academic methodologies do not always explicitly address requirements such as governance, infrastructure scalability, modularity, and integration with legacy systems. This paper examines limitations in the explicit coverage of such requirements by existing knowledge graph construction processes in compliance-driven corporate environments. We introduce the Corporate Knowledge Graph Process (CKG-P), a prescriptive, modular framework that incorporates explicit phases for governance, infrastructure planning, and continuous ontology refinement. CKG-P was empirically evaluated through a real-world, mission-critical project at the Brazilian Attorney General’s Office (AGU), focused on asset-concealment investigation. The case study involved the integration of 15 heterogeneous governmental databases containing more than 3 billion source records using an OLAP-to-Graph architecture, resulting in a production Knowledge Graph with 976,843,239 unique nodes and 1,197,618,013 relationships. The evaluation combines quantitative evidence from the case study with a structured process-level comparison against existing KG construction approaches. The results provide evidence of the operational feasibility of CKG-P in a large-scale, compliance-driven environment and show that the proposed process explicitly incorporates organizational requirements related to business understanding, data governance and compliance, infrastructure planning, the separation of deterministic and inference-based activities, and role allocation that are absent or only partially addressed in the reference processes. These findings provide empirical evidence for the applicability of CKG-P to enterprise KG projects facing comparable organizational and technical constraints while delimiting the comparison to methodological coverage rather than controlled experimental superiority.