Mines aggregate user behavior across web search to enhance relevance. Pre-Navboost-era click-driven ranking primitive — Microsoft's parallel investigation into behavior-derived ranking signal.
Patent Overview
- Inventor
- Eric Brill, others
- Assignee
- Microsoft Corporation
- Filed
- 2006-03-03
- Granted
- Published 2007-09-06
The Challenge
The Challenge
User behavior on search results — clicks, dwell, returns, refinements — carries strong relevance signal. Mining behavior across the user pool yields aggregate signals that enhance relevance beyond pure content/link signals.
- Behavior Reveals Real Relevance — Per (query, result), behavior reveals what users actually find useful.
- Aggregate Mining Reveals Patterns — Per query-pool, aggregate mining reveals patterns single sessions hide.
- Multi-Signal Behavior Capture — Per session, clicks, dwell, returns, refinements all carry signal.
- Mining Must Preserve Privacy — Per user, mining respects privacy.
- Pre-Navboost-Era Foundation — Per behavior-mining era, this primitive predates and influences Navboost-style aggregations.
Innovation
How The System Works
The system captures multi-signal behavior per session, aggregates across user pool, mines patterns per query, enhances relevance via aggregate behavior signal, and respects privacy throughout.
- Capture Multi-Signal Behavior — Per session, clicks, dwell, returns, refinements captured.
- Aggregate Across Pool — Per query, aggregations across user pool.
- Mine Patterns — Per query, patterns mined.
- Enhance Relevance Signal — Per (query, result), enhanced signal computed.
- Apply In Ranking — Per query, enhanced signal modulates ranking.
- Privacy Preserve — Per user, signals handled with privacy.
- Continuous Mining — Per fresh data, mining continues.
Behavior Mining Enhances Relevance
The patent's load-bearing idea is that aggregate user behavior — mined across the pool — produces relevance signal stronger than content/link signals alone.
Multi-Signal Behavior Aggregation
Per session, multi-signal behavior captured. Per query-pool, aggregation reveals patterns.
- Multi-Signal Capture — Per session, multiple behavior signals.
- Pool-Wide Aggregation — Per query, aggregated across pool.
- Pattern Mining — Per query, patterns mined for signal.
Technical Foundation
Technical Foundation
The patent specifies the behavior capturer, aggregator, pattern miner, signal enhancer, ranking integrator, and privacy layer.
- Behavior Capturer — Per session, multi-signal capture.
- Aggregator — Per query, pool-wide aggregations.
- Pattern Miner — Per query, patterns mined.
- Signal Enhancer — Per (query, result), enhanced signal.
- Ranking Integrator — Per query, signal modulates ranking.
- Privacy Layer — Privacy safeguards on signals.
The Process
The Process
Capture runs continuously; mining runs on rolling windows; ranking application runs per query.
- Capture Behavior — Per session, captured.
- Aggregate — Per query, aggregated.
- Mine Patterns — Per query, mining runs.
- Enhance Signal — Per (query, result), enhanced.
- Cache — Per (query, result), cached.
- Apply Ranking — Per query, ranking modulated.
- Refresh — Per fresh data, refresh.
Quality Control
Quality Control
Wrong behavior mining damages relevance. The patent specifies safeguards.
- Privacy Preservation — Per user, signals handled with privacy.
- Manipulation Detection — Per pattern, manipulation flagged.
- User-Pool Diversity — Per query, diverse user-pool required.
- Pattern Validation — Per pattern, validation against ground truth.
- Continuous Recalibration — Models refresh.
Real-World Application
Behavior-mining underpins click-driven ranking across modern engines. The pattern of multi-signal aggregation plus pattern mining informs modern Bing ranking and was the conceptual predecessor to Navboost-style systems.
- Multi-signal Behavior Capture — Clicks, dwell, returns, refinements.
- Pool-aggregated Mining Scope — Aggregations across user pool.
- Pattern-mined Signal Discovery — Per query, patterns mined.
Why Engagement Behavior Wins Over Time
Per (query, result), engagement behavior compounds. Pages earning real multi-signal engagement (click + dwell + return) accumulate ranking signal stronger than click count alone.
Why Behavior Mining Cross-Engine Compounds
Both Bing and Google reward behavior-mined signal. Content driving engagement on either engine compounds across both platforms over time.
<\/section>What This Means for SEO
What This Means for SEO
Aggregate user behavior is mined to enhance relevance — a pre-Navboost click-driven ranking primitive. SEO implication: multi-signal engagement (click, dwell, return) accumulates into ranking signal stronger than click count alone.
- Multi-Signal Engagement Compounds — Behavior mining captures clicks, dwell, returns, and refinements together. Pages earning the full engagement pattern accumulate stronger signal than click count alone.
- Satisfaction Beats Attraction — Mined behavior reveals which results genuinely satisfy. Optimize the post-click experience, not just the click.
- Aggregate Patterns Reveal Quality — Pool-wide behavior mining surfaces patterns single sessions hide. Consistent satisfaction across many users is the durable signal.
- Cross-Engine Engagement Compounds — Both Bing and Google mine behavior. Content driving engagement on either platform compounds across both over time.
- Refinement Signals Reveal Gaps — Post-click query refinements signal unmet intent. Content that fully answers reduces refinement and signals completeness.
- Privacy-Preserving Means Scale Matters — Mining requires diverse user support with privacy safeguards. Genuine audience scale produces usable behavior signal.
- Manipulation Patterns Are Flagged — Manufactured behavior is inconsistent with genuine patterns and gets filtered. Authentic engagement is the only lever.