LLMs, Content Graphs, and The Golden Age for Knowledge Graphs, from KGC2023

In early May we took part in the Knowledge Graph Conference (KGC 2023) in New York.
Here are some reflections from what we heard at the conference, plus some related insights from the Data Language team.

What is the opportunity around LLMs and Knowledge Graphs?

• There was much talk on this topic.
• We heard some echos of the hallucination issue (as seen in our KG and LLM hack event here) and also the issue of "keeping LLMs on the rails" where they just would NOT follow instructions.
• However, aside from all the hype and [well-deserved] amazement around LLMs, there is consensus that there is a practical and valuable use case for them alongside knowledge graphs.
• Specifically, LLMs have great promise in "getting information into a knowledge graph" and also in "Deriving intelligence from a knowledge graph"
• We explored a few opportunities in these spaces at our LLM + Knowledge Graph hack event in April.
• Take-home: Watch this space, as we expect it to develop rapidly.

Why have we not seen more "Content Graphs" in Media?

A "Content Graph" is a particular knowledge graph use case, where content (text, imagery, video, video moments, podcasts, audio, etc etc) is represented and managed within the graph, alongside the more usual structured metadata that we are used to finding in graphs in the Media sector.
Our talk on Content Graphs in the Media is here on SlideShare (and an overview of Content Graphs is here). In this talk, we outlined that there have previously been some large issues preventing adoption of Content Graphs in the media (we visited some of these in 2021 in Gartner's barriers to entry for knowledge graphs ).

Main reasons for the scarcity of Content Graphs in The Media:

1. Baggage: The amount of academic and theoretical overheads that seemed to be dragged into any knowledge graph project, making it too complex for stakeholders to understand.
2. Lack of good tooling: The tools required to launch a production grade knowledge graph were inaccessible and require expertise and experience. To boot, the tools have not previously facilitated rapid testing and iteration with stakeholders and SMEs.
3. Integration woes: In order to launch a Content Graph, a business would need to integrate several platforms (CMSs, DAMs, graph databases, et al), often across several organizational units. This seldom runs smoothly.
4. Slow to market: As a result of the three factors above, Content Graph platforms took a long time to get to market, which adds considerable risk of failure.

You can see more information on our Content Graph Platform here, which addresses all of these barriers.

Are we entering a Golden Age for Knowledge Graphs?

Reflecting on the content in the KGC talks, the chatter on socials, and mixing with colleagues, there is a sense that we are entering a golden age for knowledge graphs. Is this true?

We say "Yes", because launching a knowledge graph is faster, easier, and more affordable than ever

Our take is that the knowledge graph tooling that is now available as "Software as a Service" (SaaS) means that the creation and launching of high-value knowledge graphs is many times easier, less risky, more affordable, more fun, and faster than it ever has been.

From "1 year and £100ks" down to "one day and £1ks"

For a sense of scale, a knowledge graph platform project that just a few years ago took ~one year and >£500k to launch (with all the bumps "on the way" that long projects experience) can now be tested and proven within hours, for just a few £ks, using SaaS tools. This is a rough and broad estimate, but it highlights an order of magnitude shift that is game-changing for any project.
What's more is that you can explore and play with your knowledge graph, test it, adjust it, and prove it, in collaboration with your broad stakeholder ecosystem, really very easily. For information professionals, this has made work a lot more productive and enjoyable.

The factors behind this being the Golden Age for Knowledge Graphs

1. Knowledge graphs are now accessible: There is now great tooling available as SaaS, ready to go, so there is no need to research and pin-together complex sets of tooling yourself!
2. Far, far lower risk: There are large risks that plague projects when deliveries last months; making strategic decisions as a group, communicating and testing model ideas, preparing and agreeing API approach, discovering and absorbing unknown unknowns, changes in your market, changes in business strategy, team fluctuations, and more.
3. Lower Total Cost of Ownership: Not only are all the hard strategic tech platform decisions made, but the Knowledge Graph as a Service (KGaaS) platforms are managed, maintained and supported for you, saving you £ms in R&D costs, build costs, and support costs, hugely reducing your TCO.
4. Easy to test: Ease of knowledge graph model creation, data loading, and testing, which means there is no need to wait a year to find out if you have it right; you can iterate in minutes and test and test again!
5. Ready to go: Ease of production platform launch - no need to integrate a graph database, create your own APIs, integrate multiple DAMs, CMSs, CRMs - you can use a KGaaS platform and get started rapidly!

What next?

We are looking forward to next year's KGC!
In the meantime, if you are interested in hearing more about knowledge graph tooling and explorations around knowledge graphs and AI, sign up to our newsletter below.