Authority-ranking weight and confidence functions. Microsoft's formal authority-ranking framework — complement to Google's document-side authority work, with explicit confidence modeling.
Patent Overview
- Inventor
- Susan T. Dumais, others
- Assignee
- Microsoft Corporation
- Filed
- 2010
- Granted
- 2012-09-04
The Challenge
The Challenge
Authority signals carry confidence — some are well-established, some are noisy. Authority-ranking must model both weight (how much authority counts) and confidence (how reliable the signal is). Combined weight-and-confidence ranking produces robust authority application.
- Authority Signals Have Varying Confidence — Per signal, confidence varies.
- Weight Plus Confidence Combination — Per (signal, resource), weight applied with confidence modulation.
- Low-Confidence Signals Underweight — Per signal, low confidence reduces effective weight.
- Multi-Source Authority Combines — Per resource, multiple authority signals combine.
- Calibration Validates Combination — Per ranking, validation against engagement.
Innovation
How The System Works
The system computes per-resource authority signals, models per-signal confidence, combines weight-and-confidence-modulated signals, ranks results, and validates against engagement.
- Compute Per-Signal Authority — Per resource, per signal, authority computed.
- Model Per-Signal Confidence — Per signal, confidence estimated.
- Combine Weight And Confidence — Per (signal, resource), effective contribution = weight × confidence.
- Aggregate Across Signals — Per resource, multi-signal authority aggregated.
- Rank Results — Per query, results ranked.
- Validate Against Engagement — Per ranking, engagement validates.
- Recalibrate — Weights, confidence, combination models refresh.
Weight Plus Confidence
The patent's load-bearing idea is that authority-ranking requires both weight and confidence modeling. Low-confidence signals must underweight; high-confidence signals get full weight.
Confidence-Modulated Weighting
Per signal, effective contribution = weight × confidence. Confidence modeling robustifies authority ranking.
- Per-Signal Authority Computation — Per resource, per signal, authority computed.
- Per-Signal Confidence Modeling — Per signal, confidence estimated.
- Combined Weight × Confidence — Per (signal, resource), effective contribution.
Technical Foundation
Technical Foundation
The patent specifies the authority computer, confidence modeler, combiner, aggregator, ranker, validator, and recalibration loop.
- Authority Computer — Per resource, per signal, authority computed.
- Confidence Modeler — Per signal, confidence estimated.
- Combiner — Per (signal, resource), weight × confidence.
- Aggregator — Per resource, multi-signal authority aggregated.
- Ranker — Per query, authority modulates ranking.
- Validator — Per ranking, engagement validates.
The Process
The Process
Authority computation runs at indexing; ranking applies per query.
- Compute Authority — Per resource, per signal, authority computed.
- Model Confidence — Per signal, confidence estimated.
- Combine — Weight × confidence per signal.
- Aggregate — Per resource, multi-signal authority.
- Receive Query — Query arrives.
- Rank — Authority modulates ranking.
- Validate — Engagement validates.
Quality Control
Quality Control
Confidence modeling determines authority quality. The patent specifies safeguards.
- Confidence Validation — Per signal, confidence validated against held-out data.
- Multi-Signal Convergence — Per resource, multi-signal convergence reduces single-signal risk.
- Manipulation Detection — Per signal, manipulation flagged.
- Weight Bounds — Per signal, weight bounded.
- Continuous Recalibration — Models refresh.
Real-World Application
Confidence-modulated authority ranking is foundational to robust authority application. The pattern of weight × confidence informs modern authority systems where signal reliability varies.
- Per-signal Confidence Granularity — Each signal carries its own confidence.
- Weight × confidence Combination — Effective contribution scales with confidence.
- Multi-signal Aggregation — Per resource, multiple authority signals combine.
Why Strong Multi-Source Authority Wins
Per resource, multi-source authority with high confidence on each source produces the strongest aggregate signal. Single-source authority with low confidence underweights.
Why Verifiable Authority Compounds
Per signal, verifiability raises confidence. Authority that can be cross-verified across sources earns high confidence and full effective weight.
<\/section>What This Means for SEO
What This Means for SEO
Authority is applied as weight times confidence, so a strong signal that cannot be trusted is discounted before it influences ranking. SEO implication: build authority that is verifiable across independent sources, not just present, because low-confidence authority underweights.
- Confidence Gates Your Authority Signals — Each authority signal is modulated by how reliable it is. A claim of expertise that the system cannot corroborate gets reduced effective weight. Make your authority checkable, not merely asserted.
- Cross-Verifiable Authority Earns Full Weight — Signals that can be confirmed across independent sources score high confidence and apply at full strength. Citations, consistent entity references, and corroborating mentions raise confidence on your authority signals.
- Multi-Source Authority Beats Single-Source — Authority from several independent sources, each high-confidence, aggregates into the strongest signal. Concentrating all your authority claims on your own properties is weaker than authority distributed across third parties.
- Noisy Single Signals Get Discounted — One loud but unreliable authority signal underweights. A spike in self-referential or low-quality endorsements adds little because confidence on it is low. Quality of source dominates volume.
- Engagement Validation Closes The Loop — The system validates authority combinations against engagement. Authority that does not translate into user satisfaction loses calibrated weight over time. Pair authority-building with content that actually serves the user.
- Consistency Raises Confidence — Consistent representation of who you are and what you cover across sources reads as reliable. Contradictory or scattered identity signals lower confidence and dilute authority weighting.
- Earn Authority In The Topics You Claim — Confidence is assessed per signal and resource. Authority earned in a focused area is more verifiable than thin authority spread everywhere. Depth in a domain produces higher-confidence authority than breadth without proof.