Definition
Automatic enterprise knowledge graph construction is the process of resolving entities, relationships, documents and events from existing systems into a live, queryable graph — continuously, not as a one-off mapping project.
The problem
Organizations already store customers, contracts, policies and events. They do not store them as a shared map. Analysts reconstruct that map by hand. Agents cannot. The construction problem is not extraction. It is resolution: the same customer in CRM, billing and a PDF must become one node, with provenance on every edge.
What exists today
- Manual ontology modeling and expert workshops, which go stale as soon as the organization changes.
- Catalog and lineage tools that describe datasets, not organizational meaning.
- One-shot LLM extraction over documents, which produces a graph that cannot be trusted or maintained.
Deca10 research question
Can an organizational knowledge graph be constructed continuously from heterogeneous enterprise sources, with enough resolution quality that people and agents can query it as infrastructure rather than as a prototype?
Evidence
Published construction benchmarks against live enterprise systems are in progress. This page states the problem and the evaluation bar: entity resolution quality, edge provenance coverage, and time-to-usable graph from connected sources.