Christian, together with the authors and contributors of the ECLASS working group, you recently published an important Technical Specification for ECLASS as RDF. The process was complex, what were the challenges?
First, it is important to understand that we did not start from scratch. ECLASS exists for over 20 years and is used by many companies in a wide variety of systems and use cases. This means that we had to find a solution that accurately fits to the ECLASS Standard and not change it. The transformation depends on the use case and there are many visions. We decided on a universal transformation because there is no right or wrong, only good or bad. In the end, we agreed to transform ECLASS with all the subtleties, albeit complex ones, to enable a direct transfer of existing data to RDF.
Can you give us an overview of the knowledge of RDF (Resource Description Framework) and how it is used in today‘s technology landscape?
Knowledge graphs offer a wide range of approaches. Among other things, they are used to describe, store and, above all, evaluate complex data structures, dependencies and correlations. There are things that are difficult or impossible to implement with other technologies. This does not mean that knowledge graphs are easy, but the framework provides the appropriate methods and tools. RDF is a standardized framework for implementing knowledge graphs. In the standardization environment of ECLASS, the use of RDF as a W3C standard was obvious.
RDF can be used to model knowledge and relationships. In the context of products, this can be the characteristics of a product description, but also the relationships between components or assemblies in the case of physical products – as well as relationships to the installation location, etc.
Based on RDF as a framework for describing these relationships using triples, the graphs can be exchanged in different formats (e.g., Turtl), stored in special databases, queried using special queries (e.g., SPARQL), and validated using shapes (e.g., SHACL).
What are the next steps?
The currently published specification primarily describes the exchange of the ECLASS dictionary as RDF. This is the foundation for everything that builds upon it. The solutions based on it provide added value. Further parts are planned for storing specific rules in SHACL shapes, creating instances or querying data.
What challenges do you see in using RDF in practice, especially with regard to data integration and querying?
The topic is often new territory for companies or employees. Larger companies are already working on solutions or using these technologies, or there are (small) highly specialized solution providers in this area. The main challenge is to integrate this technology into the existing system landscape. Data needs to be migrated and systems need to be adapted. It is not enough to change the database. The queries are simply different. Before this can happen, however, the appropriate basic models - many speak of core ontologies - must be defined for the domain or the company. The ECLASS RDF approach provides a building block for such core ontologies.
What companies are using the technology today?
Unfortunately, I can't say anything about that. But the authors of the document certainly don't do it as a hobby. (laughs)
There are also more and more solutions and service providers offering this technology. The ECLASS subsidiary BCON² GmbH, among others, has accompanied the development and realized the first ECLASS RDF transformation, which can be used by BCON² customers directly in the browser.
What innovative applications or projects can be realized with RDF?
There are many. A major application is data management and integration. Different data sets can be linked using RDF / Knowledge Graph. Data can also be annotated and then queried accordingly.
What does the future of RDF look like as technologies like the Internet of Things (IoT) and Artificial Intelligence (AI) evolve?
In the context of IoT or Digital Twin, I see huge potential for this technology and ECLASS RDF. The W3C Web of Things Thing Description was also the motivation to start work on ECLASS RDF. There are also efforts to map the Asset Administration Shell to RDF. Using web standards for IoT and using RDF as a layer in the Semantic Web, enriched with unique URIs from the ECLASS dictionary, seems very reasonable to me.
In addition, AI can then reach its full potential using the standardized and already structured data. I can also imagine AI being used to generate complex queries from natural language. AI will reduce complexity on the application side. In principle, however, AI and RDF or graphs are complementary.
AI can help with the application of ECLASS, especially classification and description, and enable partial automation. This is very important because classification and subsequent description is a time-consuming manual process. If AI can make suggestions here - and this is possible, ECLASS itself has already evaluated this - AI will help enormously. In addition to fine-tuning large language models, concepts such as Retrieval Augmented Generation (RAG) based on vector search methods (a kind of semantic search) are used. However, only small excerpts are ever compared. This approach misses the big picture. This is where an extension to graph-based RAG solutions comes in, which also takes contextual knowledge into account. From ECLASS' point of view, this is also worth an additional evaluation.
