From Taxonomies over Ontologies to Knowledge Graphs

April 12, 2016
As in previous years, PoolParty is sponsor of the SEMANTiCS conference. SEMANTiCS conference has put the industry usages of Linked Data and the Semantic Web always into the center. So, we took the opportunity to talk with Andreas Blumauer, Product Owner of PoolParty, about the nexus between 'well-shaped' enterprise ontologies, knowledge graphs, and the successful implementation of a corporate semantic web.
 
SEMANTiCS: PoolParty Semantic Suite supports companies worldwide in implementing enterprise knowledge graphs. Can you tell us about the transition effects in such companies that start to build their own corporate semantic web?
 
Andreas Blumauer: A company that becomes dedicated to developing its own corporate semantic web has most probably gone through a lot of painful data integration and information migration projects. Therefore it has already reflected in detail how important it would be to have agile tools and methodologies in place to link information across data silos. 
 
Basically there are three types of enterprise linked data projects: Firstly, companies that develop enterprise knowledge graphs from a top-down approach, secondly, from a bottom-up perspective, and thirdly when involving third-party data or even ontologies that are used for whole industries.
 
The first kind of project can only succeed when in parallel also activities on departmental level take place. A "Corporate Semantic Web" per se is an abstract concept. If no concrete benefits can be shown quickly, in most cases such projects will be shut down even before the implementation has been started. 
On the other hand, ontologies, enterprise taxonomies, metadata harmonisation, etc. will only unfold its potential when being used on a larger scale. Using standards-based metadata and knowledge models only on a departmental level doesn't generate enough benefit to justify the overhead that is mainly triggered by the steeper learning curve when using rather new technologies like RDF graph databases. 
 
So it seems that first of all a hybrid approach, a middle-out approach is promising. For instance, when two or three departments can share efforts and costs by using the same ontologies and taxonomies, maybe even for linking some of their data sources in a more meaningful way. The benefit of this third kind of approach becomes even more obvious when industry-wide accepted ontologies or taxonomies like FIBO, MeSH, or schema.org come into play. It's all about network effects.
 
SEMANTiCS: An own ontology is a living asset for the company. It grows and continues to gain importance with every usage that is connected to it. Can you describe, which requirements you see for a "perfectly fitting" enterprise ontology?
 
Andreas Blumauer: As you've described it: ontology and taxonomy management is most often an ongoing process, thus maintainability depends on the cost-benefit ratio of it. Enterprise knowledge graphs usually consist of various ontologies, taxonomies and all kinds of structured and unstructured data transformed into RDF. 
 
We use ontologies and taxonomies for at least two reasons: they play an important role for a more precise linking of information, but also to provide more sophisticated ways to index and query data. For the first purpose, in many cases, there is no need for complex ontologies. It's also important that an enterprise knowledge graph uses ontologies that are divided into modules and several layers of abstraction like upper ontologies and lower ontologies (as we've all learned at university :-)
 
The challenge in real-world projects is that, of course, many different stakeholders with various skills levels will contribute to it. And those who will work on the taxonomical level and who will describe specific domains will have completely other expectations regarding the resulting applications than stakeholders working rather on the interoperability layer. So, a perfectly fitting enterprise ontology is designed to serve the needs of all of the users. This also means that it should develop complexity over time in parallel to the knowledge engineering skills of involved people.
 
We've frequently seen that a good starting point could be to develop first a SKOS taxonomy. This knowledge model can then, step-by-step, extended and enriched by ontologies and linked to other RDF data. PoolParty Semantic Suite is centred around this particular knowledge modelling approach - From Taxonomies over Ontologies to Knowledge Graphs.
 
SEMANTiCS: And as a final question: What is your experience on how enterprises shape their ontologies in possible relation to other companies and the market. Are there criteria which shouldn't be missed?
 
Andreas Blumauer: We live in the age of attention economy. This is valid for any asset, i.e. also for ontologies. Beside the most basic principle that existing ontologies should be reused wherever appropriate, any addition or changes should be made accessible, visible, and comprehensible for as many stakeholders as possible. 
 
Some ontology projects tend to deliver the most perfect ontology from an engineering perspective, but unfortunately forget about developers who usually prefer rather pragmatic approaches like schema.org or JSON-LD. Some ontology projects failed solely due to the fact that no marketing, training or any kind of transfer has been made. Linked Open Data has become popular most probably because of its LOD diagram that just shouts out loudly: it's a lot and you can link it!
 
Some other ontologies have become popular because of its simplicity, others still struggle with low adoption rates although heavily pushed by influential industry associations. In many cases the key success factor is accessibility, which brings me back to the standards-based approach provided by the Semantic Web combined with its scalable complexity and expressiveness (From SKOS to OWL).
Well-shaped ontologies that are designed for the reuse by other companies on the market provide various access points to serve the needs of different stakeholders with varying interests and skills levels.
 
It's like with maps: you can zoom in and out. Depending on the zoom level you will see specific types of things. Each ontology is multi-dimensional and multi-layered and an interface must be designed carefully since it's yet another detail in a bigger whole.

  • PoolParty is Gold Sponsor at this year's SEMANTiCS Conference.
    Visit the booth in the exhibition space!
  • Video: PoolParty Semantic Suite in a Nutshell (2:41 minutes)
  • Andreas Blumauer is CEO of the Semantic Web Company and Product Owner of the PoolParty Semantic Suite.
    Do not miss his talk on Enterprise Ontologies & Knowledge Graphs in this year's main programme.