Knowledge Graphs - Computerphile

Knowledge graphs help computers make sense of data for search, recommendations and other AI tasks, Professor Elena Simperl of Kings College London explains.


This video was filmed and edited by Sean Riley.

[edz]

take a shot everytime she says Bush house

[briancoverstone]

Our company processes unstructured CV data to make it structured. Obvious if you're in the HR industry. We were sued by IBM for patent infringement on processing "unstructured CV data" 10 years ago, when we had prior art years before their patent. Just ridiculous and left a disdain towards IBM. With the advent of AI and machine learning, hopefully this type of trolling doesn't occur as often.

[circuitgamer]

This was very interesting to learn about the data structure itself, but is there a good source (preferably video, but the format doesn't matter too much) to learn about how a knowledge graph is populated in the first place? That's a pretty vital bit of information that I don't know.

[PerLundholm]

Great to see that kind of paper she drew on, still exists! Haven't seen it in decades. 😀

[timangus]

If anyone is interested in visualising graphs, I wrote an application called Graphia for this purpose. Its main use is in biological sciences, but it's designed to be general purpose.

[MeppyMan]

About 30 years ago I used to setup a database system called Vineyard that came from Finland I think. It was all about building relationships between objects graphically. It let you discover interesting relationships and connections you couldn’t otherwise see.

[pankajkhushalani]

EXACTLY WHAT I NEEDED!
Had been looking into ontology and KGs for the past few days for an NLP project and I am going for it :)

[evanbarnes]

Just want to put in a plug for Obsidian Notes if anyone wants to play with building their own knowledge graph! Just started using it and it's really cool

[shemmo]

as person who works with Tigergraph and Neo4j i like this video

[fslurrehman]

Is there an open source code that creates knowledge graph 📉 out of a given data? No matter whether a text, pics or video or tables?

[andybaldman]

My god, I am in love with the way she says ‘knowledge graphs’.

[MK-jekz]

This method seem to suck when you look something fringe stuff that has more popular "synonyms". In those cases it's not uncommon that Google completely ignores a word or two from the search to steer it toward something more popular.

[lukejohnson]

How do companies like Google and others that have huge amount of this data make sure that they don't get two or more edges with different labels that reference the same data? For example, let's the underlying relationship is "location", but for LSE the edge is labelled "located in" and for KCL "location"? If that were the case they couldn't effectively query it anymore right.

[jesseparrish]

just started playing around with Dgraph/GraphQL. Dgraph has an entirely free cloud starter if anybody wants to tinker. (Not for those unfamiliar with querying databases.)

[EmptyChordMusic]

I'm not quite sure this explanation works well enough without explaining a bit more about ontologies in the first place. I've been working with RDF and related stuff for quite a bit in an enterprise PLM context, and I wouldn't have seen a knowledge graph as an AI building block - but maybe my thinking is a bit off... I might also have mentioned a few of the upsides and downsides of practical work with a knowledge graph, but I guess this is a bit too specific for an intro video.

[rshnewton]

Nicely explained and pleasant to listen to. Can you take this further for us? ♡

[jackyman]

would like to know whether this is also called a data model. if not what are the differences between a data model and a Knowledge Graph?

[itforall]

It's been always interesting to listen intellegence people talking about computer science

[johnmarianhoffman]

I work in medical AI, but I’m not particularly familiar with knowledge graphs, and this explanation didn’t really clarify anything about what they are or how they work, or the difference between data vs. “knowledge” in this context.

Just me? Anyone else just feel like we talked about Bushhouse for twelve minutes, and a very surface level concept of mathematical graphs?

[mrtienphysics]

where do you still get those dot-matrix printer papers?