1 Wait, What The Heck Is A Knowledge Graph?
I can give you a bunch of academia and business definitions of knowledge graphs with detailed explanations and use cases and you will still be puzzled and without a clue about what exactly is a knowledge graph and how it works. A very recent explanation I found in the wonderful paper Generations of Knowledge Graphs: The Crazy Ideas and the Business Impact by Luna Dong[1] – Principal Scientist at Meta Reality Labs, leading the ML efforts in building an intelligent personal assistant[2], reads:
“Knowledge Graphs (KGs) have been used to support a wide range of applications, from web search to personal assistant […] KGs model the real world in a graph representation, where nodes represent real-world entities or atomic (attribute) values, and edges represent relations between the entities or attributes between entities and atomic values. A piece of knowledge can be considered as a triple in the form of (subject, predicate, object), such as (Seattle, located_at, USA). The data instances in a KG follow the ontology as the schema, which in itself is represented in a graph form and can be taken as a part of the KG. The ontology describes entity classes, often organized in a hierarchical structure and also called taxonomy, and meaningful relationships between classes.”
As you can see, knowledge graphs are really complex technology with lots of moving parts and new concepts to grasp, given you, like me, are a non-technical person, without data modeling, data harmonization and entity linking knowledge and practical experience. And although there has been a great uptake in marketing content (me being part of that effort and guilty as charged[3]) that explains and promotes knowledge graphs to wider audiences, knowledge graphs do remain something that takes time and background knowledge to fully grasp.
As a starter, I chose to share with you a very accessible illustration by prof. Elena Simperl of King’s College London. Asked to explain knowledge graphs and the nodes and edges they are built of, prof. Simperl grabbed a piece of greenbar computer paper and drew a node to denote Bush House, further drew an edge and then connected the Bush House to a picture from a brochure lying on the table[4].
And this is the simplicity at the end of complexity when it comes to linking data from heterogenous sources. Knowledge graphs are this – a structure to represent connected entities from different sources, expandable and easy to link one to another. Just like prof. Simperl did. In the video you will see how she builds on top of the graph, taking from what’s around – it’s like she puts her cognitive semantic network to paper, literally, to a graph. Drawing a line to an image from a newspaper, she is then naming the connections and expanding the representation. In technical terms that might well translate into having one schema – ontology which defines the entities you want to describe in your graph and then taking from the outside