Is an ontology required to create a knowledge graph or is the schema derived from the relationships in a graph. What I am wondering about is whether the predicate is defined in a knowledge graph, and if so, how? It seems that knowledge graphs can be created without defining relationships between the nodes in advance.
Ralph Poole ,
Putting aside the definition of what is a knowledge graph as is discussed in [1]. My answer would be no, you do not need an ontology. But that being said you need at least a resource of frames on which to process the data. One such possibility is Verbnet. By using verbnet I've been able to derive sematic triples to use in my work.
Hope this helps
[1]Ehrlinger, L., & Wöß, W. (2016). Towards a Definition of Knowledge Graphs. SEMANTiCS (Posters, Demos, SuCCESS), 48(1-4), 2.
Well, lets be clear about the definition of the term "ontology" first. Follwing Tom Gruber "An ontology is a specification of a conceptualization." (drawn from http://www-ksl.stanford.edu/kst/what-is-an-ontology.html) "That is, an ontology is a description (like a formal specification of a program) of the concepts and relationships that can exist for an agent or a community of agents." Borst defined an ontology as a “formal specification of a shared conceptualization” (from https://iaoa.org/isc2012/docs/Guarino2009_What_is_an_Ontology.pdf), in order to be precise about the meaning that the community of agents should agree on the conceptualisation. Studer, Benjamins & Fensel extended this by the term "formal", but we can ignore this aspect for the moment.
Anyway, a knowledge graph - put aside how it is defined - should represent knoweldge about the world. Without any ontological definition of concepts (of which the nodes in the graph are instances) and the relations beween them, a knowledge graph is nothing more than a semantic network for which the criticism of Woods "What's in a link?" applies, i.e. he criticed that the links in a semantic network lack a proper meaning.
This can be seen more or less as the starting point of research on description logics, which by the way build the foundation of OWL. An important contribution of this research was the distinction of T-Box and A-Box, i.e. the terminological and the assertional component of knowledge representation. In that sense the T-Box represents an ontology following all definitions above. The point now is that the T-Box specifies our conceptualisation and hence at the same time fixes the terminology we use to describe a domain of discourse. The A-Box contains the contingent knowledge of the domain of discourse. In other words, if our domain of discourse is information about the world then a knowledge graph to large parts belongs to the assertional component and is described in terms of the terminology of the T-Box.
That are the reasons why I would say: if a knowledge graph should express knowledge (and not only information) about the world an ontology is an essential precondition for sharing the knowledge in the KG between agents.
Thomas Hoppe ,
The question as posed is that of requirement i.e. necessity. If the question was aimed at just a single interpretation of knowledge graph it would state is an ontology required under X interpretation of knowledge graph.
In having a reliance on an ontology there is an assumption that it captures all the nuances from the extracted source. Note that in your post you mention that "community of agents should agree on the conceptualisation", this is a prescriptive definition not a descriptive definition. It is an imposition for the analysis not a description of what is in the source.
Further, on analysis of sources, it is necessary to differentiate between the types of sources. If the assumption is that the analysis the analysis should not discriminate the source (and obviate analysis of the source), then it is necessary to assume I.I.D. sources and no information of the source and how to contextualize the source. Upon an I.I.D. source, the context comes from the source itself. Imposing an ontology as per the definition quoted above from your post will lead to error in the analysis of the document.
knowledge graph = ontology + data fully loaded together inside of a database).
If a KG was not built based on a previous modeled formal ontology, entity types, entity properties and entity relations can be defined based on an implicit schema.
As new data is added, types, properties and relations can be discovered and added to the schema, in a schemaless (or schema later) approach.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs