What on Earth is a Knowledge Graph? (and related relevant questions)

If you have been reading the Linked Digital Future report or any other literature about digital discoverability, you may have stumbled upon the term “knowledge graph” and wondered what on earth it can be.

Even experts disagree as to what a “knowledge graph” actually is. In simple terms, one could say that a knowledge graph is the combination of two things:

1. A data model (a conceptual model for representing information as data, with formal ontologies providing a set of rules about how knowledge must be organized within a given knowledge domain); and,
2. The actual data, stored in a graph database.

This ‘simple’ definition involves a fair deal of technical jargon. More than one could explain over a single post. So let’s focus on the second part of it: “data, stored in a graph database”.

So, what is a graph database?

According to Wikipedia, “a graph database is a database that uses graph structures for semantic [i.e., meaningful] queries with nodes [i.e., data unit], edges [i.e., relationships or links], and properties to represent and store data.”

There is a lot to unpack in this definition, but the most important aspect to understand is the edges – or the relationships – that connect data units to one another. Graph databases hold the relationships between data as a priority. Without relationships, there can be no graph database, no linked data and no knowledge graph.

Here’s an example:

— Christine Beaulieu is a cast member in “J’aime Hydro”.

If this sentence were a graph, Christine Beaulieu and the documentary theatre production J’aime hydro would be two nodes. And “is cast member in” would be their edge or relationship.

Let’s remove that relationship from the sentence. We are left with:

— Christine Beaulieu. J’aime Hydro.

The two nodes still exist. However, without their relationship, they are just sentence fragments that no longer carry any meaning.

Now, we could change that relationship to “is creator of” and we would end up with a different meaning associated with the same two nodes. These two nodes can also have many other connections with other nodes. For this reason, graph databases resemble very much social networks. Social networks can be multifaceted and intertwined: the same person can be a colleague to several co-workers, a parent to three children, and a friend to other people. In the same fashion, J’aime hydro can be multiple things to different people: a creation project for a theatre company, a work contract for an actor, a touring show for a presenting organization, and a live performance for a theatre-goer.

To simplify all of the above, one could say that the knowledge graph and the graph database are about relationships and meaning. Very much like social networks. Or the performing arts value chain.

But why is any of this important for the arts sector?

A graph database is fundamentally different from a traditional relational database. Their structures are different. They enable different ways of using and sharing data. And they can lead to different ways of thinking about data.

Let’s have a comparative look at the relational database and the graph database.

Relational database	Graph database
Relationships between tables (for example, a list of organizations and a list of venues) are implicit by virtue of the structure of the database.	Relationships are explicit: the connection between two nodes denotes a specific relationship and this relationship is a data point in the graph database.
Linear. Handles one-to-one and one-to-many relationships between records very well. Has a harder time with many-to-many relationships.	Multi-directional. Excels at many-to-many relationships. Any node can have as many relationships with other nodes as you can think of. A node can be one thing and multiple other things at the same time (just like human beings).
Good for protecting data. A relational database can nonetheless exchange data with another database via an API.	Good for exposing and exchanging data. A graph database can denote objects and relationships using the Resource Description Framework, a standard W3C model for data interchange on the Web of data. RDF-based graph databases can be exposed as linked open data, and linked to other RDF databases.
Good for scaling vertically (adding more of the same type of data), but cannot easily scale horizontally (adding different types of data).	Good for scaling horizontally. If new needs/use case arise, one can easily expand a graph database to accommodate new types of data (and serve new use cases).
Relational database can provide training data for machine learning.	Graph database makes machine learning possible (i.e. machine learning involves identifying patterns and inferring new relationships). Graph theory is the foundation of AI.

As the above table illustrates, relational databases and graph databases are not only fundamentally different in the way they organize data; they enable very different digital possibilities. And they can lead to different digital mindsets.

One is about sorting things in buckets. The other is about valuing relationships between things.

My bet is that the latter is more apt at generating the kind of radical collaboration that the arts sector needs to fully realize its digital transformation.

The author thanks Gregory Saumier-Finch, CTO at Culture Creates, and Jai Djwa, Principal at Agentic Communications, for their contributions to this post.

Recommended readings

MT Buzzer, Graph database vs. relational database, July 26, 2018.

Favio Vázquez, Graph Databases. What’s the Big Deal?, January 22, 2019.

Stefan Summesberger and Juan Sequeda, Knowledge Graphs Need Social-Technical Solutions, May 24, 2019.

Josée Plamondon, Web sémantique : de choc culturel à transformation numérique, 16 juillet 2018.

Josée Plamondon, Produire des données : entre outils de marketing et bases de connaissances, 21 août 2019.