Entity & Knowledge Graph Patents Vintage
The shift from keyword-based search to entity-based search represents the single largest evolution in how Google understands the web. Bill Slawski tracked this transformation through dozens of patent analyses, documenting how Google moved from matching strings of text to understanding real-world things and their relationships.
The Evolution from Keywords to Entities
How the Knowledge Graph Updates Itself
One of Bill's most important discoveries was a patent describing how Google's Knowledge Graph grows autonomously by answering questions.
The Self-Updating Knowledge Graph Process
Key Patent Insights
- The Knowledge Graph identifies gaps in its own data — missing attributes, relationships, or facts about entities
- It generates natural language questions to fill those gaps
- Answers come from multiple sources: web content, structured data, crowdsourced answers
- Confidence scoring determines whether an answer is added automatically or flagged for review
- The system is designed to be self-improving — each answer can reveal new gaps to fill
"The future of search is in providing knowledge to searchers through a Knowledge Graph. Web pages connected by links are information points connected by nodes. This was the first generation of web search. The Knowledge Graph represents the next generation." — Bill Slawski's analysis (2018)
Source: How the Google Knowledge Graph Updates Itself by Answering Questions (2018)
Entity Extraction from Web Content
Google patents describe multiple methods for identifying entities in web content:
Methods of Entity Extraction
| Method | Description | Patent Era |
|---|---|---|
| Named Entity Recognition | Identifying proper nouns (people, places, organizations) | 2008-2012 |
| Co-occurrence Analysis | Entities that appear together frequently are related | 2010-2014 |
| Structural Analysis | Entities found in lists, tables, and structured HTML | 2012-2016 |
| Fact Extraction | Pulling subject-predicate-object triples from text | 2014-2018 |
| Automated Data Wrappers | Programs that extract structured data from semi-structured pages | 2016-2020 |
Fact Answers: How Google Extracts Facts from Pages
When Google crawls the web, it does not just index words. It extracts facts as structured triples:
Google has been working on object-level search since at least 2005, moving from page-level retrieval to extracting and ranking individual objects (products, people, papers, organizations) embedded within pages.
Source: Providing Fact Answers (2017)
Knowledge Panel Triggering
Bill analyzed several patents related to how Knowledge Panels appear in search results.
What Triggers a Knowledge Panel
Based on patent analysis, Knowledge Panels appear when:
- An entity is recognized in the query as a known Knowledge Graph entity
- Entity attributes are available (description, images, related entities)
- Confidence threshold is met — the system is confident about which entity the user means
- Query intent is informational and entity-seeking (not navigational or transactional)
Entity Disambiguation
When a query matches multiple entities, Google uses:
- Context from the query — additional words that narrow the entity
- User location — geographic context for local entities
- Search history — prior queries that suggest which entity variant the user means
- Popularity priors — the most commonly searched entity gets preference
Co-Occurrence and Concept Research
Bill's analysis of the "Not All Anchor Text is Equal" concept revealed how Google uses co-occurrence (words and entities appearing near each other) as a ranking signal.
Co-Occurrence vs. Co-Citation
| Concept | Definition | SEO Implication |
|---|---|---|
| Co-occurrence | Two entities or concepts mentioned near each other on a page | Establishes topical relevance between concepts |
| Co-citation | Two pages linked from the same third page without linking to each other | Establishes relationship between pages/sites |
| Anchor text | The clickable text of a hyperlink | Describes the target page in the linker's words |
The Shift from Keyword Research to Concept Research
Bill frequently wrote about how the entity-based approach meant SEOs needed to think about concepts rather than just keywords:
- Keyword research: What exact phrases do people type?
- Concept research: What entities and relationships exist around a topic, and how can content demonstrate comprehensive understanding?
This shift means creating content that demonstrates topical authority by covering the full entity graph around a subject — related entities, attributes, relationships, and contextual information.
Source: Not All Anchor Text is Equal and Other Co-Citation Observations (2012)
User-Specific Knowledge Graphs
One of Bill's later analyses (2021) covered a patent for on-device, user-specific Knowledge Graphs used to rewrite mobile queries.
How User-Specific Knowledge Graphs Work
- Built from data across multiple apps on a mobile device
- Stores personal entity information (contacts, locations, preferences)
- Used to rewrite queries with personal context
- Operates on-device for privacy — not sent to Google servers
- Provides personalized search results that understand "my dentist" or "my office"
Source: Rewritten Queries and User-Specific Knowledge Graphs (2021)
Key Takeaways
- Entities are the new keywords — Google thinks in terms of real-world things and their relationships, not just text strings.
- The Knowledge Graph grows itself — It actively identifies gaps and fills them, meaning entity coverage will only increase.
- Co-occurrence builds entity associations — Mentioning related entities together on a page strengthens topical relevance.
- Structured data feeds the Knowledge Graph — Schema markup is a direct input into entity recognition systems.
- Personal knowledge graphs are coming — Search is becoming personalized at the entity level, not just the query level.
- Concept coverage beats keyword density — Comprehensive coverage of an entity's attributes and relationships signals expertise.