Knowledge Graph Deep Dive Vintage

The Google Knowledge Graph, launched in 2012, represents Google's shift from a search engine that finds pages to one that understands things. Bill Slawski tracked its development through multiple patent analyses, revealing the mechanisms behind how Google builds, populates, and maintains this massive entity database.

Knowledge Graph Architecture

How Google Populates the Knowledge Graph

Bill identified multiple data sources and mechanisms through patent analysis:

Data Sources

Source	Type	Example
Wikipedia / Wikidata	Structured reference	Entity attributes, descriptions
Freebase	Structured database	Entity relationships (acquired by Google)
Web crawling	Semi-structured extraction	Facts from web pages
Schema markup	Structured data	Entity data from website owners
Google Maps	Location data	Business entities, geographic entities
Books	Text extraction	Historical and reference data
User queries	Implicit data	Entity attributes inferred from questions

The Fact Extraction Pipeline

Source: Providing Fact Answers (2017)

Self-Updating Knowledge Graph

Bill's 2018 analysis revealed one of the most fascinating Knowledge Graph patents — a system that identifies its own gaps and fills them.

The Self-Improvement Loop

How Gap Identification Works

The system identifies gaps by:

1. Comparing entity completeness — If most entities of type "Person" have a "birth_date" but some do not, those missing dates are gaps
2. Analyzing query patterns — If users frequently ask "When was X born?" for entities without birth dates, that confirms the gap
3. Cross-referencing sources — If structured data provides attributes that the Knowledge Graph lacks, those are gaps
4. Monitoring freshness — If an entity's attributes have not been updated recently, they may be stale

What This Means for SEO

• Provide complete entity information on your website — the Knowledge Graph actively looks for missing data
• Use structured data to make entity attributes machine-readable
• Answer common questions about your entity — these answers may be harvested for the Knowledge Graph
• Keep information current — the system detects stale data

Source: How the Google Knowledge Graph Updates Itself by Answering Questions (2018)

Entity Reconciliation

One of the hardest problems in building a Knowledge Graph is reconciliation — determining when different references point to the same real-world entity.

The Reconciliation Challenge

Reconciliation Signals

Signal	How It Helps
Context words	Surrounding text disambiguates ("Apple iPhone" vs. "apple pie")
Structured data	Schema markup with unique identifiers (SameAs URLs)
Co-occurring entities	"Tim Cook" + "Apple" = Apple Inc., not the fruit
Source type	A tech blog mentioning "Apple" likely means the company
Geographic context	"Apple" + "Cupertino" = Apple Inc.

Question Answering to Populate the Knowledge Graph

Bill's 2014 analysis covered a patent describing how Google uses question-answering to fill Knowledge Graph gaps.

The Q&A Pipeline

1. Identify incomplete entities — Find entities with missing attributes
2. Generate questions — Create natural language questions for the missing data
3. Search for answers — Query the web for pages that answer these questions
4. Extract answers — Use NLP to pull specific answers from text
5. Verify answers — Cross-reference with multiple sources
6. Update the Knowledge Graph — Add verified answers as new entity attributes

Missing vs. Incorrect Data

The patent also addresses detecting and correcting incorrect data in the Knowledge Graph:

• Conflicting sources — When multiple sources provide different answers, the system analyzes source authority
• Temporal changes — Facts that change over time (CEO of a company, population of a city) need updates
• User feedback — Reported errors trigger verification processes

Source: Question Answering to Populate the Knowledge Graph (2014)

Entity Authority and Trust

Not all entity sources are treated equally. The Knowledge Graph uses trust signals to determine which sources to rely on:

Source Trust Hierarchy

Key Takeaways

1. The Knowledge Graph is self-improving — It identifies its own gaps and fills them automatically.
2. Entity reconciliation requires consistency — Use the same name, identifiers, and SameAs links everywhere.
3. Structured data feeds the Knowledge Graph directly — Schema markup is not just for rich results; it is a data input.
4. Answer common entity questions on your site — Google harvests answers to fill Knowledge Graph attributes.
5. Source authority matters — Being referenced by high-trust sources improves the quality of your entity data.
6. Keep entity data current — The Knowledge Graph detects stale information and seeks updates.