Entity Linking (Reconciliation)

Entity linking is the step that connects a mention of a name to the specific entity it refers to in a knowledge base. Entity extraction finds that “Apple” is an organisation in a sentence; entity linking decides it is Apple Inc. and attaches its identifier, such as a Wikidata item or a Knowledge Graph ID. When the same idea is applied to whole datasets, matching your records against an external knowledge base, it is usually called reconciliation.

How it works

A typical linking pipeline has three steps:

• Mention detection: find the names in the text (named entity recognition).
• Candidate generation: look up every knowledge base entry that could match, for example all entities called “Mercury”.
• Disambiguation: pick the best candidate using context. Surrounding words, co-occurring entities, the page’s topic and its schema all help.

Tools expose the result in different ways. Google’s Natural Language API returns metadata with a Knowledge Graph MID and a Wikipedia URL when it can link an entity. The Knowledge Graph Search API lets you look up an entity and see its ID and a confidence score. OpenRefine can reconcile a spreadsheet column against Wikidata.

At web scale, Google keeps its Knowledge Graph current through a reconciliation process described in its patents: new facts are clustered from different sources, matched to existing entities using key facts and added as new nodes or relationships when no match exists.

How it’s used in SEO and marketing

• Entity audits: check whether your brand, people and products are linked to a Knowledge Graph ID when your pages are analysed. No ID often means Google does not confidently recognise the entity.
• Keyword and content research: link extracted entities to Wikidata to pull in types, attributes and related entities.
• Structured data: sameAs links and a stable @id make reconciliation easier by stating which external records describe you.
• Data cleaning: reconcile brand, product or location lists against a reference source before analysis.

Example

A page about a jazz musician called Mercury mentions albums, a record label and a tour. Extraction finds “Mercury” as a person. Candidate generation returns the planet, the element, the Roman god and the musician; the co-occurring music entities push the linker to the musician’s Wikidata item. Without that context, the mention could be linked to the wrong entity or left unlinked.

Limitations and things to watch

• Small or new entities often have no knowledge base entry, so they cannot be linked yet.
• Linkers make mistakes on ambiguous names, and confidence scores vary between tools.
• Inconsistent names across your site and profiles make reconciliation harder.

Learn it on MLforSEO

• What Is a Knowledge Graph? How to Use It for Smarter Keyword Strategy
• How to Audit Entity Structure in Modern Search
• Entity Disambiguation in Practice
• Knowledge Graph API Extraction and Entity Relationship Analysis (Notebook)
• The AI-powered Entity Audit