A knowledge graph is a way of storing information as things and the links between them, so software can follow those links to answer questions. Each thing is a node, such as a product or a customer, and each link is an edge, such as "bought" or "belongs to".
How a knowledge graph is built
A graph stores facts as small statements of the form thing, relationship, thing:
| Thing | Relationship | Thing |
|---|---|---|
| Ad 14 | belongs to | Ad set "Spring buyers" |
| Ad 14 | uses | Creative "Studio photo" |
| Creative "Studio photo" | shows | Product "Trail shoe" |
| Customer 805 | bought | Product "Trail shoe" |
Link these statements together and you can ask a question that crosses several tables in a normal database: which creatives showed products that customers then bought? A spreadsheet would need several lookups. A graph follows the links directly.
Let Madgicx's AI do the work
Its AI marketer finds what's working and prepares the next move. You approve.
Try for Free ($0 Trial)Knowledge graph versus database
| Relational database | Knowledge graph | |
|---|---|---|
| Shape | Tables of rows and columns | Nodes and relationships |
| Strong at | Counting, summing, filtering | Following connections across many types of things |
| Adding a new kind of link | Often a new table | A new relationship type |
| Meaning | Held in column names | Held in the relationships themselves |
Search engines use knowledge graphs to connect facts about people, places and brands. AI systems use them to ground answers in connected facts. Our entry on machine learning covers the models that sit on top.
A knowledge graph for Meta ads
A Meta ad account is already graph-shaped. A campaign holds ad sets, ad sets hold ads, ads use creatives and audiences, creatives point to products in a catalog, and results attach to each level. Our answer on campaigns, ad sets and ads lays out the first layers.
An advertiser can extend that graph with data of their own: orders, first-party data, custom audiences and creative traits such as hook, format and offer. The payoff is questions a flat report can't answer, such as which creative traits win in which audience.
This is the advertiser's own data structure. Meta's published detail about its systems is narrower: its engineers describe Andromeda, which narrows "tens of millions of ad candidates into a few thousand relevant ad candidates". Meta's public pages don't go into how its delivery systems store relationships between ads, so read claims about Meta's internal graph with care.
How to start a small one
- Pick the question. For example, which creative traits lead to purchases.
- List the node types. Ads, creatives, audiences, products, orders.
- Define the relationships. Uses, targets, shows, bought.
- Load data with stable IDs. The same ad ID must mean the same node everywhere.
- Add labels. Tag creatives with traits such as format and offer. Creative analytics can supply them.
- Query and test. Check that answers match the numbers in Ads Manager.
Example. A sneaker store labels 40 ads with their creative, product and audience, then links each product to its orders. The graph shows that 18 of its 40 ads feature the same trail shoe, yet only 3 of those 18 reached a repeat buyer. The team reviews the other 15 and rebuilds their creative for the shoe's returning buyers.
In Madgicx. AI Marketer audits your Meta ad account across campaigns, ad sets, ads and creatives and shows what drives results, and Creative Insights breaks down every creative so the traits behind winners are visible. The AI prepares the work and you approve it.
FAQ
What is a knowledge graph in simple terms?
It is a network of facts. Each thing is a dot, each relationship is a line between dots, and software follows the lines to answer questions.
What is an example of a knowledge graph?
A graph linking products, customers and orders, so you can ask which products customers who bought X also bought. Search engines use much larger ones to connect facts about people and places.
How is a knowledge graph used in marketing?
It links data that usually sits in separate tools, such as ads, creatives, audiences and orders. That lets a team ask questions across them.
Does Meta use a knowledge graph for ads?
Meta's published engineering material on its ad systems, such as Andromeda, focuses on retrieval and ranking, and the source used here doesn't describe a knowledge graph. Advertisers can still build one from their own account data.
Is a knowledge graph the same as AI?
No. A knowledge graph is a way to organize data. AI methods such as machine learning can use it, but it works without them.
Related
- Related terms: Machine learning, First-party data, Creative analytics
- In depth: Knowledge graphs for Meta ads
Sources
- Meta Andromeda retrieval engine, Engineering at Meta, December 2024
