Knowledge graphs: a friendly definition
You may or may not have heard the term “knowledge graph” used recently, especially in relation to AI or companies like Google, IKEA, Electronic Arts, Microsoft, and Facebook; for example, Netflix’s graph-based recommendation engine being worth $1 billion a year, those panels on the side of Google results pages, or when Mark Zuckerberg wanted to “reset Facebook” by “wiping everyone’s graph“. But what exactly is a knowledge graph, when are they useful, and why are they so important in an AI-obsessed world?
[/et_pb_text][et_pb_image src=”@ET-DC@eyJkeW5hbWljIjp0cnVlLCJjb250ZW50IjoicG9zdF9mZWF0dXJlZF9pbWFnZSIsInNldHRpbmdzIjp7ImVuYWJsZV9odG1sIjoib2ZmIn19@” _builder_version=”4.18.0″ _dynamic_attributes=”src” _module_preset=”default” hover_enabled=”0″ global_colors_info=”{}” sticky_enabled=”0″][/et_pb_image][et_pb_text admin_label=”Google Trends” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”]
Gartner’s Generative AI (GenAI) Impact Radar for 2024 listed two technologies as being “Now” (immediately impactful): GenAI itself, and knowledge graphs (KGs). GenAI as “Very High Mass” of impact and knowledge graphs as “High”.
Yet, while business networks and the mass media are flooded with talk of GenAI, knowledge graphs get little discussion outside niche circles.
[/et_pb_text][et_pb_image src=”https://urbinaconsulting.com/wp-content/uploads/sites/7/2026/05/Gartner-impact-radar-2024.jpg” title_text=”Gartner-impact-radar-2024″ _builder_version=”4.27.4″ _module_preset=”default” hover_enabled=”0″ global_colors_info=”{}” alt=”Gartner Impact Radar for 2024 graphic showing that GenAI and knowledge graphs are the most potentially impactful technologies right now.” sticky_enabled=”0″][/et_pb_image][et_pb_text admin_label=”part 3″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”]Semantic relationships support human relationships
Imagine a giant spider web. At each connection point, there’s a bit of information: maybe a fact about a person, a place, a thing, or even an intangible concept. These bits of information are connected by threads in the web that represent the relationships between them. This is what a knowledge graph is like. It’s a vast, interconnected network of facts and how those facts relate to each other.
A knowledge graph is a special type of database that understands things in a natural, human way – by their meaningful real-world concepts and their relationships – rather than data tables and rows.
What is a “human, natural way” to relate data? To illustrate, try to think of random words, starting from the jumping off point “James Bond”.
You’ll find you come up with a list something like:
- Movie
- Action
- Thriller
- Book
You might have thought of specific characters or prominent elements like:
- Sean Connery. Roger Moore, or Daniel Craig: actors who played the character
- Moneypenny: another character from the franchise
- Aston Martin: A type of car James bond frequently drove
And so on. You might have also thought of synonyms for the above, like “film” instead of “movie”.
True “randomness” is extremely difficult for the human brain. We tend to think in these related webs of concepts and facts. Knowledge graphs are built in exactly this way: the architecture of the database mirrors human webs of meaning. For this reason, we call them a “semantic” (meaning-based) technology.
[/et_pb_text][et_pb_image src=”https://urbinaconsulting.com/wp-content/uploads/sites/7/2024/04/Semantic-relationships-support-human-relationships-v2-scaled.jpg” title_text=”Semantic relationships support human relationships-v2″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][/et_pb_image][et_pb_text _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”]Knowledge graphs in action
If you do a Google search for “Daniel Craig”, Google’s may pull information showing that Daniel Craig is an actor who starred in movies like Casino Royale and Glass Onion: A Knives Out Mystery. It would also link him to related entities like wife Rachel Weisz and other high profile actors like George Clooney or his children.
Google uses a KG to display useful snippets of information prominently in search results so you get intuitive answers faster. You can imagine how the concepts of Movie, Genre, Year, Cast, and their corresponding datapoints Casino Royale, Action/Thriller, 2006, and so on map onto various types of related results pages.
[/et_pb_text][et_pb_image src=”https://urbinaconsulting.com/wp-content/uploads/sites/7/2024/04/Daniel-Craig.jpg” title_text=”Daniel Craig” _builder_version=”4.27.4″ _module_preset=”default” hover_enabled=”0″ global_colors_info=”{}” sticky_enabled=”0″][/et_pb_image][et_pb_image src=”https://urbinaconsulting.com/wp-content/uploads/sites/7/2024/04/casino-royale-delivery.png” title_text=”casino royale delivery” _builder_version=”4.27.4″ _module_preset=”default” hover_enabled=”0″ global_colors_info=”{}” sticky_enabled=”0″][/et_pb_image][et_pb_text _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”]When someone (or a computer system) in a company looks for information, a knowledge graph helps them see not only the piece of data they’re looking for but also how it connects to other relevant information. The web shows you both where you are on it, and also the interesting places nearby and the paths you can take to get there.
When you combine the subject domain concepts and data – in this case movies – to business data, then you can create powerful applications. This is why companies like Netflix use it for mapping personal user data to their graph in their recommendation technology (warning, very techie article!), which as far back as 2016, were saying was worth $1 billion to them, per year according to Yahoo Finance.
Knowledge graph use cases and benefits
- Finding connections: They help uncover relationships between pieces of information that might not be obvious at first glance. For example, linking customer feedback to specific product features or identifying common issues in customer service inquiries.
- Automatic tagging: Because they understand the core concepts relevant to a given domain, knowledge graphs can drive a system that reads millions of documents or images and tags them appropriately according to what is in the KG.
- Recommendation: Recommendations are a multi-faceted power. You can recommend anything from related content to recommending “Next best action (NBA)” or “next best engagement (NBE)” – which means automatically surfacing insights or content to customers or staff to take the action they see fit. All this can be driven via intranets, extranets, apps, or even the metaverse based on the best available contextual data.
- Enhancing search: When you search for something in a company’s system, a knowledge graph can help by understanding the context of your search and bringing back more relevant results.
- Improving decision-making: By providing a comprehensive view of the company’s data and how it all relates, knowledge graphs can support better, more informed decision-making.
- Personalization: For businesses that interact directly with customers, knowledge graphs can help tailor experiences, recommendations, and services to individual needs and preferences based on the interconnected data.
- Innovation: By making it easier to see how different pieces of information relate to each other, knowledge graphs can spark new ideas and ways of thinking about problems or opportunities.
In simple terms, a knowledge graph is like a super-smart librarian who knows not only where every book is but also how each book relates to every other book in the library. It’s a powerful tool for businesses to organize, search, and leverage their data in ways that were previously impossible or very difficult to achieve.
Inside large companies, knowledge graphs power many intelligent services by understanding facts and relationships at scale. From personalized content recommendations (e.g. Facebook uses one) to smart searching and question answering (e.g. Google and Amazon Alexa use them), knowledge graphs are the brains behind many modern AI applications. They allow these services to connect dots between people, places, things and ideas so they can better serve their users.
How do knowledge graphs work in an enterprise tech stack?
In a business setting, this spider web can stretch across the entire company (or just across closely related systems), connecting all sorts of data. Ideally, it could be data from different departments – sales, customer service, marketing, product development, and more. In practice, it’s often related departments who have a specific business need to connect what they do for internal or external users, or even AIs.
Knowledge Graph are gaining traction in the enterprise technology stack because they address key gaps in traditional table-based relational systems:
Artificial Intelligence: Because they hold a “ground truth” understanding of a knowledge domain, its concepts, data, and all the interrelationships amongst them, KGs are an ideal back-end to keep generative and other types of AI delivering explainable, reliable results.
Flexible Connectivity: KG’s bring a new level of data and content connectivity, improving how businesses can organise and deliver content across multiple channels.
New Applications: From enhancing search functionalities to powering recommendation systems and auto-taggers, knowledge graphs are at the heart of creating intuitive, human-centric, scalable digital experiences.
Integrated Digital Experiences: As we move towards more integrated digital landscapes, the role of knowledge graphs in creating meaningful and context-aware links across various systems is becoming increasingly critical.
How do knowledge graphs support AI?
A study by data.world saw that knowledge graphs can vastly improve AI accuracy and performance.
[/et_pb_text][et_pb_image src=”https://urbinaconsulting.com/wp-content/uploads/sites/7/2024/04/data_word_kg_study.svg” alt=”Study showing the superiority of knowledge graphs vs regular databases” title_text=”data_word_kg_study” _builder_version=”4.18.0″ _module_preset=”default” global_colors_info=”{}”][/et_pb_image][et_pb_text _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”]A benchmark study by data.world compared how well ChatGPT 4.0 could answer questions using two different approaches: querying a traditional database (using something called SQL – structured query language) versus using a knowledge graph, with human-meaningful (semantically structured) relationships.
Accuracy results: KGs vs traditional databases
- 71.1% vs 25.5% for simple questions that required finding straightforward facts
- 66.9% vs 37.4% for complex questions requiring interpretation and insight from multiple related facts
- 35.7% vs failing entirely at 0% for simple questions that required combining many related facts to get a complex answer
- 38.7% vs 0% when both the question and answer were complex (interestingly, slightly higher than for simple questions)
Knowledge Graphs lower the chances of AI hallucination
Generative AIs, like Large Language Models (LLMs, like ChatGPT, Claude, and Gemini), are not databases. They excel at generation or manipulation tasks like summaries, translation, and drafting. They are weak on data management or accurate retrieval. They are very useful, but still unreliable, because they return plausible-sounding but incorrect information. The common term for this is to “hallucinate”.
See also: “The Unreliable Computer Revolution” – Noz Urbina, LinkedIn and “Are AI models doomed to always hallucinate?“, Kyle Wiggers, TechCrunch
Pairing a language AI with KGs means that the GenAIs stick to what they do best, generating and communicating, and leave the data accuracy and domain understanding to the knowledge graph. AI becomes just a layer over the data.
[/et_pb_text][et_pb_text _builder_version=”4.18.0″ _module_preset=”default” global_colors_info=”{}”]Leveraging knowledge graphs on the web and beyond
In summary, knowledge graphs are special databases that hold a structured web of facts that help computers broadly understand the human world.
By encoding millions of interrelated facts, they enable applications to surface relevant information and insights faster.
So, next time you get a useful recommendation or a quick answer from a digital system – be that a website, app, or even an in-game link, or maybe inside virtual reality – there may well be a knowledge graph behind the scenes.
If you’ve got a question or would like to learn more about implementing knowledge graph technology in your business, get in touch.
[/et_pb_text][et_pb_text disabled_on=”on|on|on” _builder_version=”4.27.4″ _module_preset=”default” disabled=”on” global_colors_info=”{}”]
to do
- Add something about why this is better, more scalable than dbs
- Zuckerberg quote
- Stats
- Add descriptive line inside images so they’re self-contained
[/et_pb_text][/et_pb_column][/et_pb_row][/et_pb_section]