Why Ontology Is Suddenly in China’s National Data Policy and What It Means for AI

On June 3, 2026, the National Data Administration issued the “Implementation Plan for Promoting High‑Quality Industry Data Set Construction,” outlining six special actions and dozens of tasks. The author notes that the term “ontology” appears in the document for the first time in an official Chinese policy.

“Strengthen the construction of knowledge bases, knowledge graphs, ontologies and other data sets, accelerate the construction of datasets for complex task planning, long‑term reasoning, human‑machine interaction, decision execution, and empower new intelligent application forms such as agents.”

Ontology (Ontology) is a lower‑level concept than knowledge graphs. The author compares three data concepts:

Database – metaphor: warehouse – core function: tells you which shelf an item is on.

Knowledge graph – metaphor: map – core function: tells you how different stores are connected.

Ontology – metaphor: blueprint – core function: defines what exists in the world and the essential relationships.

In AI, ontology is a metadata structure designed for machines: it defines entities such as “fund,” “manager,” and “FOF,” and their logical relationships, enabling models to truly understand rather than merely match strings.

Why now?

The author identifies three urgent drivers:

Breaking the statistical bottleneck of large models : Current large models are massive text compressors that struggle with complex logical reasoning and industry knowledge. Ontology offers a structured way to inject business logic into AI.

Fuel for evolving agents : As AI shifts from chat to action, agents need an engine that understands business rules. Without an ontology, an agent is like an autonomous car without a map.

Qualitative shift from “managing data” to “using data” : After years of building infrastructure, the national signal is that the data flywheel must start turning, and the core step is semantic preparation.

Personal relevance

The author reflects on reading the policy while working on the Dify project. In a previous product‑question‑answer assistant, the model often gave wrong answers because the knowledge fed to it lacked a skeleton. In a recent investment‑director agent, the author deliberately applied ontology modeling:

Define entities: fund, fund manager, custodian, …

Define relationships: manage, custody, distribution, …

Define attributes: net asset value, fee, key indicators, …

The national push for industry‑level ontology standards means that the second half of data governance will move from “interconnection” to “mutual understanding,” representing a true qualitative change.

Conclusion

The author reiterates a previous claim that the next decade of data governance will shift from “controlling data” to “making data understandable.” Those who can perform ontology modeling and understand knowledge‑graph engineering will possess the “foundational language” of the AI era, creating a hidden professional advantage.