ZTE's O-R-R-A Ontology Framework: Making Agents Understand Business Semantics

Problem: Agents Cannot Understand Business Despite Decades of Data Governance

After 20 years of data governance, BI tools and AI agents still fail to comprehend business semantics: metric definitions are inconsistent, entities are misaligned across systems, and AI hallucinations persist. Data integration solved connectivity, but the "understanding" layer remains broken.

Root Cause: Four Layers of Systemic Semantic Interoperability Gaps

The bottleneck is not model capability but insufficient semantic supply. The article identifies four missing layers:

Missing semantic annotations

Cross-system concept heterogeneity

Non-formalized business rules

Undefined operational semantics

Solution: 璇玑 DataAgent + O-R-R-A Ontology Framework

O-R-R-A Framework

The framework defines machine-understandable, reasonable, and executable business cognitive contracts through four primitives:

Objects – business entities

Relations – connections between entities

Rules – formalized business logic and constraints

Actions – executable operations

Incremental Ontology Engineering (Six Stages)

Scenario-driven

Identify objects

On-demand modeling

Connect data

Application validation

Continuous precipitation

This approach drives continuous growth of enterprise cognition rather than producing static documentation.

Six Runtime Components

The ontology becomes runtime infrastructure via:

Modeling Workbench

Data Mapping Engine

Semantic Query Engine

Unified Metric Service

Agent Access Gateway

Governance Console

Quantitative Results

Deployed across 10+ mainstream systems

Minute-level heterogeneous field alignment

Metric definition unification rate improved from 62% to 95%

Risk path identification time reduced from 3 days to minutes

90-day iteration cycle; MVP ontology bounded by 50 core concepts

Governance targets: cross-scenario reuse rate ≥ 60%, alignment accuracy ≥ 95%

Key Insight: Ontology is not a drawn knowledge graph, but a business cognitive contract executable by agents. The only test standard: has it entered the agent's runtime call chain?

Practical Takeaways

Diagnostic Framework: Four-layer semantic interoperability gap diagnosis and six-layer semantic enablement ladder assessment to evaluate enterprise intelligent construction stage and gaps.

Low-Risk 90-Day Roadmap: Select one high-value use case, build MVP ontology within 50 concepts, integrate one agent scenario, with three-party review and SemVer semantic versioning governance to avoid "big-bang modeling" failures.

Three Reference Implementation Patterns:

Cross-system semantic alignment with conversational analysis

Risk reasoning along relationship chains

Scenario-driven precise data governance

Each includes selection comparisons, governance mechanisms, and quantified effect metrics.