To know is to learn

A knowledge graph can be thought of as a search engine on steroids, though its application goes well beyond that to chatbots, product recommenders and autonomous systems. Semantics make it possible for the knowledge graph to carry out probabilistic analysis of content: ranking meaning against search criteria on the basis of salience referenced against various attributes & relationships. It organises content ontologically, using a rules engine to infer logical consequences (knowledge) from declared facts. Because nodes, relationships & attributes are equally important, it is easy to plumb in new datasets or shift perspective to explore for new knowledge. Knowledge graphs are flexible (e.g. supported data types or schemas) which is why they are capable of evolving to reflect changes in content or domain.

You may choose to bespoke a knowledge graph (with the help of domain experts) or adapt a pre-existing one. There are advantages & disadvantages to both approaches, depending upon your situation. Of the pre-existing knowledge graphs, Google’s is perhaps the best known, followed by Amazon’s product graph. Both are specific to the organisation which created them. Wikidata and Geonames are among those graphs which are general and openly-available. Or these may simply be too broad for your requirements, in which case, it may be worthwhile to build your own, with the help of subject experts.

Graphs, unlike search engines, assume that a search terms may have more than one interpretation, and that these can be ranked e.g. to determine which information the user was really looking for. So it’s not just about searching for documents that contain the word ‘apple’ but about the machine understanding the nature of ‘apple’ through seeing that entity in its potential context, with attributes, using applied logic that enables the knowledge graph to distinguish between an eatable apple and an Apple product. It is that richer information which enables disambiguation, something the search engines, even the fuzzy ones, never quite cracked.

Google’s knowledge graph “currently contains more than 500 million objects, as well as more than 3.5 billion facts about and relationships between these different objects”. With such eye-watering numbers, the design and technical architecture of Google’s graph must have been a major consideration in the effort to make sure that a graph of that size could still perform at scale. Google’s graph also delivers a summary, ideal for a quick look to determine whether you are on the right track, and the ability to make lateral connections and find out related but new facts that relate to the target item.

The Amazon graph learns how we buy, specifically product – it is, after all, the ‘shop of everything’.

Google’s graph learns how we think.

What do you want to learn?