Module 1: Semantic Search Origins

Foundation 2 Patents 30 min

The foundation of everything Google does in search today started with these two patents.

Patents Covered

#	Patent	Inventor	Year	Key Innovation
1	US 6,678,681	Sergey Brin	2000	DIPRE Algorithm
2	US 6,678,681B1	-	2004	Semantic Search

---

Patent 1: DIPRE Algorithm

Full Title: Information Extraction from a Database Patent Number: US 6,678,681Inventor: Sergey Brin Assignee: Google LLC / Stanford University Filed: March 9, 2000 Granted: January 13, 2004

The Breakthrough

DIPRE (Dual Iterative Pattern Relation Expansion) is Google's FIRST semantic search invention. It extracts structured data from unstructured web content - the foundational technology that became the Knowledge Graph.

How DIPRE Works

The Algorithm Explained

Step 1: Seed with Known Facts Start with known tuple pairs. In Brin's example: author-book pairs like ("Isaac Asimov", "The Robots of Dawn").

Step 2: Find Occurrences Search the entire web for where these pairs appear together. Record the surrounding context.

Step 3: Identify Patterns Analyze HOW these tuples appear. A pattern consists of:

• Prefix text - what appears before the data
• Middle text - what separates the data elements
• Suffix text - what follows the data
• URL patterns - where this pattern appears

Step 4: Extract More Tuples Use discovered patterns to find NEW tuples that match the same patterns.

Step 5: Iterate Repeat until you've extracted all matching information or reached diminishing returns.

Real Example from the Patent

Starting with 5 science fiction books:

Author	Book
Isaac Asimov	The Robots of Dawn
William Gibson	Neuromancer
...	...

Result: 15,000+ books extracted with 95% accuracy.

Why This Matters for SEO

Key Insight #1: Entity Co-occurrence

When your brand consistently appears alongside relevant entities in the same patterns, Google builds stronger associations.

Key Insight #2: Consistent Formatting

Use consistent patterns when mentioning entities. If you write "Author: John Smith" on one page, don't write "By John Smith" on another.

Key Insight #3: Structured Data Foundation

This is WHY schema markup works - it provides clean, consistent patterns for Google's extraction systems.

Practical Application

DO:

• Maintain consistent entity formatting across your site
• Co-locate related entities (author + book, product + brand, person + organization)
• Use structured data to make patterns explicit
• Create "seed" content that establishes your entity relationships

DON'T:

• Use inconsistent naming (sometimes "Dr. Smith", sometimes "John Smith MD")
• Scatter related entities across disconnected pages
• Assume Google will figure out relationships without patterns

---

Patent 2: Semantic Search

Patent Number: US 6,678,681B1

This is the conceptual extension that moves from keyword matching to meaning understanding.

The Shift

Old Approach	New Approach
Match keywords	Understand meaning
Count word frequency	Analyze relationships
Exact string matching	Semantic similarity
Documents as word bags	Documents as concept graphs

How It Changed Search

---

Module Quiz

Question 1: What does DIPRE stand for?

Answer: Dual Iterative Pattern Relation Expansion

Question 2: What three text elements define a DIPRE pattern?

Answer: Prefix text, Middle text, and Suffix text

Question 3: Why does consistent entity formatting matter?

Answer: DIPRE-style systems identify entities by recognizing consistent patterns. Inconsistent formatting makes pattern recognition harder, weakening entity associations.

---

Key Takeaways

1. Patterns enable extraction - Consistent formatting helps Google identify your entities
2. Co-occurrence builds relationships - What appears together, becomes associated together
3. Iteration expands knowledge - One connection leads to discovering more
4. Accuracy matters - False patterns create noise; precision beats volume

---

Next Steps

Continue to Module 2: Phrase-Based Indexing →

This module introduces Anna Patterson's revolutionary work that determines how Google measures topical relevance.