Definition
The process of mapping extracted text mentions to unique, canonical entries in a Knowledge Graph or structured database to resolve ambiguity. In RAG pipelines, it enables grounding by ensuring the LLM retrieves facts about the correct specific entity, though it introduces trade-offs between retrieval precision and the computational latency of graph lookups.
Distinguish from Named Entity Recognition (NER): NER identifies that a word is a person; Entity Linking identifies exactly which database record that person corresponds to.
"A digital GPS pin dropping onto a specific coordinate on a map after a user types a generic landmark name."
Conceptual Overview
The process of mapping extracted text mentions to unique, canonical entries in a Knowledge Graph or structured database to resolve ambiguity. In RAG pipelines, it enables grounding by ensuring the LLM retrieves facts about the correct specific entity, though it introduces trade-offs between retrieval precision and the computational latency of graph lookups.
Disambiguation
Distinguish from Named Entity Recognition (NER): NER identifies that a word is a person; Entity Linking identifies exactly which database record that person corresponds to.
Visual Analog
A digital GPS pin dropping onto a specific coordinate on a map after a user types a generic landmark name.