OpenBG Core Ontology

• The OpenBG core ontology is a formal framework that structures billions of business facts into a unified knowledge graph with hierarchical classes, concepts, and relations.
• It organizes business entities using semantic web standards and taxonomic hierarchies, enabling precise interoperability and context-aware analytics in e-commerce.
• The ontology integrates multimodal data and employs a phased methodology for scalable development, ensuring consistent, enriched business intelligence.

A core ontology within the OpenBG business knowledge graph provides the formal framework by which billions of business facts, spanning products, consumption scenarios, and multimodal signals, are structured and semantically unified. Serving as the foundational schema, it enables interoperability, precise taxonomy, and the integration of heterogeneous data modalities into knowledge graphs at unprecedented scale. The OpenBG core ontology draws upon established semantic web standards, incorporates design principles from situation and core ontologies in scientific applications, and is engineered for rich downstream analytics in e-commerce deployments.

1. Ontological Architecture and Formal Structure

OpenBG’s core ontology is formally defined as $$\mathbb{O} = \{\mathbb{C}, \mathbb{P}, \mathcal{R}\}$$, where:

• $\mathbb{C}$ (classes) is split into:
  • Category ($\mathbb{C}^{C}$): abstract product types (e.g., "mobile phone")
  • Brand ($\mathbb{C}^{B}$): information-rich brand entities (inc. founders, logos)
  • Place ($\mathbb{C}^{P}$): geographic and production region entities
• $\mathbb{P}$ (concepts) are simple abstractions capturing consumption demands:
  • Time, Scene, Theme, Crowd, Market Segment (aligned with SKOS roles)
• $\mathcal{R}$ (relations) is portioned into:
  • Object properties ($\mathcal{R}_{obj}$): directed links (e.g., brandIs, placeOfOrigin, relatedScene)
  • Meta-properties ($\mathcal{R}_{meta}$): axiomatic, taxonomic constraints (e.g., rdfs:subClassOf, owl:equivalentClass, skos:broader)
  • Data properties ($\mathcal{R}_{data}$): attributes describing multimodal features (e.g., imageIs for product images)

This specification enables unambiguous partitioning of entities and supports formal inferences on product-category relationships, brand provenance, and multimodal analytics (Deng et al., 2022).

2. Taxonomic Hierarchy and Semantic Interlinking

Central to the core ontology is the organization of product Classes using multi-level taxonomic relations:

• Product instances link directly to a Category via ⟨product, rdf:type, Category⟩, guaranteeing association to a well‐defined abstract type.
• Hierarchies activate via rdfs:subClassOf (for classes) and skos:broader (for concepts), e.g., a Category "mobile phone" is a subclass of the top-level owl:Thing; concepts such as "evening party" Scene may be marked skos:broader to "party".
• Object properties connect category nodes with Brand, Place, and consumption concepts such as Scene and Time, supporting compositional reasoning about when and where a product is relevant.

The structure underlies product search, recommendation, and alignment scenarios, ensuring semantic links across heterogeneous facts and enabling robust schema evolution.

3. Principle of Situation Modeling and Core Ontology Abstraction

Drawing upon layered scientific ontologies (Olsina et al., 2021), OpenBG’s schema reflects a multi-tiered architecture reminiscent of FCD-OntoArch, encompassing Foundational, Core, Domain, and Instance levels:

• The ontology’s terms are domain-independent at the Core level, enabling semantic enrichment and modular reuse.
• Entity relationships and axioms are designed for logical consistency, extending best practices from situation modeling. For example, the distinction between Target Entity (the central product in a transaction or context) and Context Entity (surrounding features such as market segment or theme) allows context-aware analytics (e.g., which products are preferred by which demographic in what season).
• Formal axioms enforce correct assignment and exclusion principles. E.g., if a product is a Target Entity in a consumption scenario, it is not a Context Entity for that assertion; if a product is influenced by a context, there exists an encompassing Situation in which both participate.

A plausible implication is that reuse of established domain-independent terms and inheritance of higher-level properties via stereotypes (<<ProcessCO::Human Agent>>, etc.) enables rapid ontology extension and alignment, critical for evolving business domains.

4. Multimodal Data Integration and Quality Control

The ontology explicitly supports the uniform representation of multimodal business data—including text, images, and tables:

• Data modalities are converted into structured RDF triples via data properties (e.g., imageIs, text comments).
• Attributes tied to products (color, weight, size) are included using minimal, high-value descriptors for efficient inference and retrieval.
• Strict enforcement of meta-properties ($\mathcal{R}_{meta}$) such as owl:equivalentClass is used to correct for noisy or inconsistent business data (e.g., duplicate or misclassified geographic names).

Scalability is achieved through ontology minimalism, reducing redundancy and maximizing maintainability, which is essential when the KG contains billions of triples and millions of entities.

5. Methodological Framework for Interoperability

Ontology engineering in OpenBG adheres to a phased ecosystem methodology (Qiang, 16 Jul 2025):

• Design phase: Ontology Design Patterns (ODPs) are applied to conceptualize recurring structures, ensuring consistent schema and rapid extension. E.g., Observation Pattern is utilized for sensor or event-driven scenarios analogous to consumption events.
• Develop phase: Ontology Matching and Versioning (OMOV) protocols align OpenBG’s evolving schema with external domain ontologies, supporting version traceability and semantic consistency. Mapping scores for alignment are computed via weighted sums of lexical, syntactic, and semantic similarities:

$$\mathrm{Score}(x, y) = \alpha \cdot \mathrm{LexSim}(x, y) + \beta \cdot \mathrm{SynSim}(x, y) + \gamma \cdot \mathrm{SemSim}(x, y)$$

• Deploy phase: Ontology-Compliant Knowledge Graphs (OCKGs) validate schema accuracy and guide pay-as-you-go refinement cycles, ensuring capture of real-world semantics and facilitating feedback-driven ontology evolution.

This phased approach is essential for ensuring cross-domain interoperability and continual schema improvement in complex, dynamic business contexts.

6. Ontology-Driven Business Applications and Impact

Through its rigorous schema, the OpenBG core ontology directly supports advanced business intelligence:

• Knowledge graph-centric tasks such as link prediction, entity extraction, multimodal summarization, and recommendation are enhanced by explicit category, brand, and context relations.
• KG-enhanced model pre-training (e.g., mPLUG variants) leverages structured multimodal knowledge, empirically improving gross merchandise volume (GMV) and click-through rate (CTR) in live e-commerce deployments.
• Uniform semantic integration of image and text boosts downstream accuracy and relevance in product matching, recommendation, and review summarization.

A plausible implication is that the explicit core ontology underlies scalable business analytics, supporting billions of triples for robust, context-aware applications.

7. Future Development, Challenges, and Standardization

• The adoption of clear semantic standards (RDF, OWL, SKOS) and modular architecture enables widespread reuse and alignment with enterprise ontologies.
• Practical challenges—such as correcting for noisy, multimodal input and maintaining large-scale consistency—are addressed through canonical constraints and iterative refinement protocols.
• The commitment to minimal but expressive schema design ensures long-term scalability and manageability.
• Ongoing development includes enhanced documentation, expanded mapping to external standards, and integration with emerging ontology frameworks such as Common Core Ontologies (Jensen et al., 27 Apr 2024).
• Community engagement via open-source benchmarks and competitions drives further validation, refinement, and application diversity.

This comprehensive core ontology architecture provides OpenBG—and similar large-scale business KGs—with a principled, interoperable, and semantically rich foundation for advanced analytics and intelligent decision-making in e-commerce and beyond.