Entity Linking
Also called Entity Linking
Figuring out which real-world thing a name in the text actually refers to.
Think of it like
Sorting mail for three coworkers all named "Jordan" — same name on the envelope, three different desks.
Example
A news pipeline reads "Apple cut its forecast" and links "Apple" to the company, not the fruit, so it files the story under the right stock.
How it actually works
Text is full of ambiguous mentions — "Paris," "Mercury," "Jordan" — that map to different entities depending on context. Entity linking matches each mention to a unique record in a knowledge base, using the surrounding words to disambiguate. It’s the glue that turns messy prose into structured facts, and the step where a knowledge graph gets its nodes. Errors here quietly corrupt everything downstream.
For product teams
Keeps your data clean at the source — "Mercury the planet" doesn’t end up in the chemistry bucket.
For engineers
Map surface mentions to canonical entity IDs in a reference KB, disambiguating with contextual features; feeds graph construction.
Related
- Knowledge Graph — Its nodes populate this structure.
- Embeddings — Uses meaning-based matching under the hood.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome