Improves search rankings by combining a topical relevance score with a location score derived from the geographic data associated with both the query and the candidate documents.
Patent Overview
- Inventor
- Amit Singhal
- Assignee
- Google LLC
- Filed
- 2003-09-22
- Granted
- 2009-10-20
- Application Number
- US 10/665,720
The Challenge
Topical Relevance Alone Misses Local Intent
A query for "pizza" from a user in Boston should rank Boston pizza places higher than New York pizza places, even though both are equally topical. Pure topical ranking has no way to know which Boston pizza place the user actually wants because it does not read location signals at ranking time. The system needs an explicit location score that combines with the topical score to bias rankings toward documents relevant to the user’s geographic context.
- Topic Without Place Is Half An Answer — Many queries have implicit location intent. Local services, restaurants, shops, and events are all worse off when the engine ignores geography. The user expects nearby results even when they did not say "near me".
- Documents Have Location Too — Web pages can be associated with locations through contact info, address mentions, IP, schema markup, or assigned regions. The engine needs to read these signals and use them in ranking.
- Need A Score That Combines With Topic — Location should not replace topical relevance; it should adjust it. The ranking needs an explicit combination function that lets location-sensitive queries weight location heavily while keeping topic dominant for non-local queries.
- Locations Are Imprecise — User location can come from IP, account profile, device GPS, query terms, or session signals. Each has different precision and confidence. The system must handle uncertain location inputs gracefully.
- Cross-Region Queries Must Still Work — Users sometimes search for places far from where they are (travel research, relocation planning). The system cannot blindly bias toward the user’s current location; it must use query-derived location when present.
Innovation
Location Score Plus Topical Score
For each candidate document relative to a query, the system computes a location score (how well the document’s location matches the query’s location) and a topical score (how well the document matches the query’s topic). The two scores combine into a ranking score. Documents that score well on both rank highest; documents that score well on only one are downweighted.
- Receive Query With Location Data — The query arrives with associated location data: user’s inferred location (IP, GPS, profile), any location terms in the query string, and session context like recent place searches.
- Identify Candidate Documents — Standard topical retrieval produces a candidate set of documents responsive to the query. Each candidate has its own location associations from the indexing pipeline.
- Compute Location Score — For each candidate, score how well its location data matches the query’s location data. Stronger matches (same city, same neighborhood) score higher than weak matches (same country).
- Compute Topical Score — For each candidate, score how well the document’s content matches the query’s topical intent. This is standard relevance scoring independent of location.
- Combine The Two Scores — Combine location and topical scores into a single ranking score per candidate. The combination weights are tunable and can depend on how location-sensitive the query is.
- Rank By Combined Score — Order candidates by the combined score and return the top-k to the user. Documents that match both location and topic float to the top.
Two Scores, One Ranking
The patent’s contribution is the explicit two-score framework. Earlier systems either ignored location or applied it as a post-hoc filter. By making location a first-class scoring axis alongside topic, the system can produce continuous trade-offs between local and global relevance per query.
Location Modulates, Doesn’t Replace
Topical relevance is the primary signal. Location is a multiplicative or additive adjustment that biases the ranking toward documents in the right geography.
- Query Location — Derived from user signals (IP, GPS, profile, session) and query content (explicit place mentions). Confidence varies by source; the score weighting accounts for that.
- Document Location — From contact info, address markup, schema, IP, assigned regions, or content analysis. Each candidate document carries one or more location associations into ranking.
- Score Combination — Location and topical scores combine via a tunable function. Location-sensitive queries weight location heavily; non-local queries leave topic dominant.
Geography becomes a ranking axis, not a filter.
<\/section>Technical Foundation
What The System Reads And Computes
The framework requires location data on both sides of the ranking comparison and a combination function that decides how much each side contributes.
- Query Location Data — Location associated with the query: user-inferred location, explicit place terms, session-derived location. Multiple sources may provide overlapping or conflicting signals.
- Document Location Data — Location associated with the document: address, contact info, schema markup, server IP, content analysis. Indexed alongside topical features.
- Location Score — Measure of geographic match between query location and document location. Higher when locations are closer or more specifically matched.
- Topical Score — Standard relevance score from topical retrieval. Independent of location.
- Combined Ranking Score — The function over location score and topical score that drives final ranking. Combination weights are tunable per query type.
Quality Metrics
- Location Score — Decreasing function of distance. Same-city matches score higher than same-state which scores higher than same-country.
loc_score(D, Q) = f(distance(loc(D), loc(Q))) - Topical Score — Standard TF-IDF, language-model, or learned relevance scores. The patent does not change topical scoring itself.
top_score(D, Q) = relevance(content(D), terms(Q)) - Combined Score — Linear combination is one option; multiplicative or learned combinations also work. Weights are tunable and can depend on query type.
rank(D, Q) = w_top * top_score + w_loc * loc_score
Key Insight: Treating location as a co-equal ranking axis with topic, rather than as a filter applied after retrieval, is what allows continuous trade-offs. A query that is 30% location-sensitive can have its ranking adjusted by 30%-weight location score, rather than being forced into either a global or a local result set. The continuous trade-off matches how user intent actually distributes.
<\/section>The Process
End-To-End Ranking Path
The location-aware ranking runs as part of the standard query path. Topical retrieval produces candidates; the location component is added at scoring time.
- Receive Query — User submits a query along with available location signals (IP-derived, GPS, profile, query terms).
- Topical Retrieval — Standard topical retrieval produces a candidate set of documents responsive to the query terms.
- Lookup Document Locations — For each candidate, retrieve its location associations from the index.
- Score Each Candidate Both Ways — Compute topical score and location score per candidate.
- Combine Per Query Type — Combine the two scores using weights appropriate for the query type. Highly local queries weight location more; broad queries weight topic more.
- Rank And Return — Order candidates by the combined score and return the top-k to the user.
What This Means for SEO
What This Means for SEO
Location-aware ranking is one of the foundations of local SEO. Knowing the two-score framework changes how you think about local content, Google Business Profile signals, and geographic targeting.
- Location Signals On Your Pages Are Read Explicitly — Address, contact info, structured data (LocalBusiness, Place, PostalAddress), and explicit place mentions all contribute to the document location score. Treating them as decoration misses the ranking lever.
- Be Specific About Place — Same-city match scores higher than same-state. Pages targeting local intent should name the city explicitly in titles, H1, and structured data. Vague regional targeting ("serving the greater area") weakens the location score.
- Query-Derived Location Beats User-Derived — When the query itself names a location, that wins over the user’s current location. A user in Boston searching "pizza in New York" gets New York results because the query location is explicit. Target pages by query intent, not just by user location.
- Topical Score Still Dominates Non-Local Queries — For queries with no location intent, the location score barely contributes. Don’t over-localize content that targets globally relevant queries; you waste topical authority for no location reward.
- Document Location Indexing Comes From Multiple Sources — The engine reads location from address fields, schema, IP, content analysis, and external signals (Google Business Profile). Multiple aligned sources reinforce the location signal; conflicting sources weaken it.
- Multi-Location Businesses Need Per-Location Pages — A single page that lists every city you serve dilutes the location score per city. Per-city pages, properly marked up, capture more location-aware ranking opportunities.
- Travel And Research Queries Use Query Location — When users search for places far from their own, the engine uses the query location, not the user location. Travel content can rank for destination queries from any origin if the destination is explicitly marked.
- Location Sensitivity Varies By Query Type — Restaurants, services, events, and shopping are highly location-sensitive. Information queries ("how to", "what is") are not. Match your content’s location specificity to the query type you target.