Building a Unified Ontology from a Multitude of Sources

Aim: Making machines interact with the Intelligent Enterprise.

Use Case: Assigning a source of supply to a purchase requisition by an operational purchaser.

Industry: Procurement

Overview: An ontology defines a common vocabulary for researchers who need to share information in a domain, including machine-interpretable definitions of basic concepts in the domain and relations among them. Many of the ideas for extracting ontologies, as described in this paper can be extrapolated from business processes models to any other enterprise model.

Ontologies related to an ERP context could be created from many different sources, and then enhanced and enriched with more semantics depending on the needs of the domain. Below are some potential sources of ontologies:

Knowledge of a domain expert, ex: describing a purchase order and its relations to suppliers, materials, etc.
Business process documentation, ex: creating a new Purchase Order.
Existing business process models, ex: BPMN notation.
Metadata of a RESTful APIs or services, ex: OData service metadata.
Business entity models, ex: CDS views/models, i.e. proprietary specification formats.
UI interactions ex: Fiori view/app, etc.

These ontology sources are covering an ERP Stack vertically (Process Models, Business Models, Data Models, etc.) and horizontally (covering different Lines of Business, Domains, custom domains, etc.). In order to take advantage of those ontologies which could be domain-driven but also in a different layer in the ERP stack, we would need to build a unified ontology based on a homogeneous taxonomy that creates links and relations between the entities of the respective given ontologies.

Example: NLP needs to process unpredictable and indeterministic input. User input is based on natural language which has infinite possibilities even in a business context, and for NLP to process different domains and different layers of the ERP stack, dealing with different ontologies using different taxonomies and schemas could lead to overhead and performance issues. The unified taxonomy together with the meta-ontology will serve AI solutions, such as NLP.

Tasks:

Given the examples of ontologies as starting templates and the related schemas, analyze and assess if any adaptations/enhancements are needed.
Document the assumptions and the changes of the given ontologies and the reasons behind them.
Come up with unified ontology based on one taxonomy.
Document the outcome and the reasons behind each choice.
Come up with ideas of an approach to utilize deep learning (ML) to extract this taxonomy from the given sources.

Resources: