The 'Relevance' factor for location-sensitive queries. Adjusts document ranking by how location-sensitive the query is — local-intent queries reward proximity and locale match; global-intent queries don't. Singhal co-invented.
Patent Overview
- Inventor
- Xianping Ge, Abhishek Parmar, Amit Singhal, Adam Smith, Daniel Egnor, Elizabeth Hamon
- Assignee
- Google LLC
- Filed
- 2003
- Granted
- 2012-05-01
The Challenge
The Challenge
Not every query is location-sensitive. 'Pizza' has strong local intent; 'history of pizza' does not. Ranking must read query location-sensitivity and adjust signal weights accordingly. Without sensitivity awareness, location signals over-apply or under-apply.
- Queries Vary In Location Sensitivity — 'Pizza' = high sensitivity. 'Pizza recipe' = medium. 'History of pizza' = low. The same word, different sensitivity.
- Location Signals Must Modulate — On high-sensitivity queries, proximity, locale match, business prominence dominate. On low-sensitivity queries, web-relevance signals dominate.
- Sensitivity Must Be Inferred — Per query, location-sensitivity must be inferred from query terms, click patterns, and behavioral signals. Manual labeling doesn't scale.
- User Location Context Matters — Per user, location context (current location, home location) modulates how sensitivity applies.
- Calibration Across Query Types — Sensitivity calibration must hold across query types: short-tail, long-tail, conversational, navigational.
Innovation
How The System Works
The system infers per-query location sensitivity from query patterns and click signals, weights location-anchored ranking signals by that sensitivity, applies user location context, and produces a sensitivity-aware combined ranking.
- Infer Per-Query Location Sensitivity — Per query, infer location sensitivity from query terms, click patterns, and aggregate behavioral signals.
- Capture User Location Context — Per user, capture current location and broader location context (home, work, frequent locations).
- Weight Location Signals — Per query, scale proximity, locale match, prominence signals by inferred sensitivity.
- Combine With Web-Relevance — Sensitivity-scaled location signals combine with web-relevance signals into composite ranking.
- Apply User Context — Per (user, query), user location context modulates which proximity signals matter.
- Validate Against Labels — Sensitivity inference validated against labeled query-relevance data.
- Continuous Recalibration — Inference and scaling models recalibrate against fresh data.
Sensitivity Determines Signal Weight
The patent's load-bearing idea is that per-query location-sensitivity is the gate that determines how much location signals weight ranking. Sensitivity-aware ranking applies the right signals to the right queries.
One Query, One Signal Mix
Different queries deserve different signal mixes. Sensitivity-aware ranking computes per-query signal weights rather than applying uniform weights everywhere.
- Per-Query Sensitivity Inference — Per query, sensitivity inferred from terms, clicks, behavioral signals.
- Sensitivity-Weighted Location Signals — Per query, proximity/locale/prominence signals scaled by sensitivity.
- User Location Context — Per user, location context modulates which proximity signals apply.
Technical Foundation
Technical Foundation
The patent specifies the sensitivity inferrer, user-context capturer, signal weighter, ranking combiner, validator, and recalibration loop.
- Sensitivity Inferrer — Per query, infers location sensitivity from query patterns and click signals.
- User-Context Capturer — Per user, captures location context with privacy preservation.
- Signal Weighter — Per query, scales location signals by sensitivity.
- Ranking Combiner — Combines sensitivity-scaled location signals with web-relevance into composite ranking.
- Validator — Sensitivity inference validated against labeled data.
- Recalibration Loop — Inference and scaling recalibrate against fresh data.
The Process
The Process
Sensitivity inference runs at query time alongside web ranking. Per-user context applied per query.
- Receive Query — Query arrives.
- Infer Sensitivity — Per query, location sensitivity inferred.
- Retrieve User Context — Per user, location context retrieved with privacy preservation.
- Weight Location Signals — Per query, location signals scaled by sensitivity.
- Combine With Web Relevance — Composite ranking computed.
- Return Results — Sensitivity-aware results returned.
- Capture Feedback — Click feedback feeds back into sensitivity calibration.
Quality Control
Quality Control
Sensitivity inference accuracy determines ranking quality. The patent specifies safeguards.
- Inference Validation — Sensitivity inference validated against labeled query-relevance data.
- Privacy Preservation — User location context stored with privacy preservation.
- Sensitivity Bounds — Per query, sensitivity scaling bounded. Prevents over-weighting or under-weighting.
- User-Pool Diversity — Aggregations require diverse user-pool support.
- Continuous Recalibration — Inference and scaling recalibrate against fresh data.
Real-World Application
Location-sensitivity ranking is the 'Relevance' factor in the Local Pack trinity. The pattern of per-query sensitivity inference plus user-context modulation underpins modern local search.
- Per-query Sensitivity Granularity — Each query receives its own sensitivity inference.
- Signal-weighted Application Method — Location signals scaled by sensitivity. High-sensitivity queries reward location signals; low-sensitivity don't.
- Context-aware User Modulation — Per user, location context modulates which proximity signals apply.
Why Location-Specific Content Wins For Local Intent
On high-sensitivity queries, location signals dominate. Content with clear location anchoring (service area, city pages, locale markers) earns ranking benefit on local-intent queries.
Why Mixed-Intent Pages Underperform
Pages serving both global and local intent fragment their location signal. Pages with clear local positioning (single service area, explicit locale) signal more strongly under sensitivity-aware ranking.
<\/section>What This Means for SEO
What This Means for SEO
This patent infers how location-sensitive each query is and scales location signals accordingly, so local-intent queries reward proximity and locale match while global-intent queries do not. SEO implication: clear local positioning wins local-intent queries, and mixed-intent pages fragment their signal.
- Clear Local Positioning Wins Local Intent — On high-sensitivity queries, location signals dominate ranking. Content with clear location anchoring, such as a defined service area, city pages, and explicit locale markers, earns the benefit on local-intent searches.
- Do Not Mix Global And Local Intent On One Page — Pages serving both global and local intent fragment their location signal. A page with single, explicit local positioning signals more strongly under sensitivity-aware ranking than a do-everything page.
- Sensitivity Is Inferred Per Query — The same word can carry high or low location sensitivity depending on phrasing, so pizza and history of pizza are treated differently. Target the local-intent phrasings deliberately rather than assuming a keyword is always local.
- Low-Sensitivity Queries Reward Web Relevance — On global-intent queries, web-relevance signals dominate and location is down-weighted. For informational topics, depth and authority matter more than location anchoring, so do not over-localize global content.
- User Context Modulates Proximity — The system applies the user's location context to decide which proximity signals matter. Genuinely serving the areas where your likely searchers are located is what lets proximity work in your favor.
- Sensitivity Scaling Is Bounded — Per-query scaling is bounded to prevent over- or under-weighting location. There is a ceiling on how much pure location anchoring can do, so it complements rather than replaces relevance and quality.
- Clicks Recalibrate Sensitivity — Click feedback feeds back into sensitivity calibration. Pages that genuinely satisfy local intent reinforce the local interpretation of the queries they win, which rewards real intent-matching.