What a knowledge graph is

A knowledge graph is a database in which information is stored as nodes and edges. Nodes are objects
— products, categories, brands, attributes — and edges are named relationships between them.

(Nike) --[manufactures]--> (Air Max 90)
(Air Max 90) --[belongs to]--> (Sneakers)
(Air Max 90) --[material]--> (Leather plus mesh)
(Air Max 90) --[suited for]--> (Urban style)
(Air Max 90) --[bought together with]--> (Nike Everyday socks)

That structure supports reasoning: if a shopper likes urban Nike sneakers, the graph builds a path
to other urban leather sneakers — even with no explicit interaction history.

Applications in recommender systems

Enriching collaborative filtering

Classical collaborative filtering looks only at patterns of user behaviour. A knowledge graph adds
semantics — why these products are alike. Not only because they are bought together, but because
they share a brand, sit in the same price range, or are made of the same material.

KGCN (knowledge graph convolutional networks) and KGAT (knowledge graph attention networks) train
jointly on user interactions and graph structure, and report better accuracy and diversity than pure
collaborative filtering.

Supporting search

In e-commerce a knowledge graph improves semantic search: the query warm jacket for downhill skiing
is interpreted through the graph as a search over the attributes temperature rating of -15 and below
plus activity type of alpine skiing.

Cold start through a graph

Scenario Without a graph With a graph
New product in the catalogue No recommendations until history accumulates Recommended immediately through attribute links
New visitor Popular products as a fallback Personalization through demographic and contextual links
Rare product (under 10 interactions) Low-quality embeddings Stable recommendations through graph structure

Important: the quality of a knowledge graph depends directly on the quality of catalogue
attributes. Incomplete product descriptions, an inconsistent category taxonomy, duplicate brands
spelled differently — all of it reduces the graph’s value. Building one starts with ETL work to
normalise the catalogue.

Knowledge graphs in SEO

Google’s Knowledge Graph, launched in 2012, is the largest public example of the technology. It
populates the information panels in search results for brands, organisations and people, drives
entity-based SEO and feeds the answers in AI Overviews. For e-commerce brands, presence in Google’s
Knowledge Graph raises the chance of appearing in structured answers.