Building and Applying an Enterprise Relationship Knowledge Graph

Knowledge graphs bring tremendous value in the fintech era, offering reasoning ability and explainability that naturally suit financial scenarios; by extracting knowledge from data with AI techniques, they provide powerful semantic representation, storage, and inference for intelligent applications.

This article shares a practical case of building an enterprise relationship knowledge graph, illustrating the full workflow from graph construction to application, and achieving three model uses: relationship path query, ultimate controller discovery, and enterprise group detection.

The motivation includes answering why an enterprise relationship KG is needed, what data are required, how to associate entities, relations, and attributes, and presenting concrete application examples.

Flow – Preparation : Identify business requirements and perform data preprocessing. The data sources comprise internal relational databases (company registration, personnel, status, and related‑company tables) and public registration platforms. Irrelevant fields and largely empty columns are removed.

Ontology Modeling : Define two main entity types—company and person—and their direct relations. Companies have four direct relations (shareholder, investment, guarantee, branch) and persons have five (shareholder, investment, guarantee, executive, contact). Tables illustrating these relations are shown below.

The cleaned data are then imported into the constructed ontology model. A portion of the ontology schema, showing company and person nodes, relationship edges (branch, shareholder, guarantee, investment, executive, contact), and selected attributes, is illustrated in the diagram below.

Model Application – Relationship Path Query : This query discovers direct or indirect paths between any two entities, revealing hidden connections and assessing risk exposure. An example query between person P2 and company C is shown, with the algorithm diagram below.

Model Application – Ultimate Controller Discovery : Using a depth‑first traversal on shareholder relations, the algorithm computes shareholding ratios along each path, aggregates them, and identifies the top‑level shareholder whose effective ownership exceeds a threshold, thus revealing the actual controlling person.

Model Application – Enterprise Group Discovery : By aggregating control paths, the method builds a control skeleton; when a controlling entity oversees a sufficient number of legally independent companies, it is considered an enterprise group. Visualizations of discovered groups are provided below.

Conclusion : By leveraging existing structured data in a bottom‑up approach, a comprehensive enterprise relationship knowledge graph was built, enabling the three mining models—path query, ultimate controller discovery, and group detection. Future extensions could include supply‑chain, licensing, sales, and family relationships, subject to data availability.