Uses social network signals (who endorsed what, who connected to whom) to personalize search results, surfacing content recommended by trusted social connections and identifying experts within the social graph relevant to the user's queries.
Patent Overview
- Inventor
- Marc Najork, others
- Assignee
- Google LLC
- Filed
- 2011-09-30
- Granted
- 2015-02-03
- Application Number
- US 13/250,929
The Challenge
The Challenge
Social network signals (endorsements, shares, follows) carry strong personalization signal but require careful integration with search ranking. Surfacing content because a friend liked it adds value; doing so badly breaks user trust. The system needed graceful integration.
- Social Signals Are Underused In Ranking — When a user has social connections that engaged with content, that engagement is relevance signal for the user. Ignoring it wastes a strong signal.
- Friend Endorsement Beats Generic Authority — A trusted friend's endorsement of an article often outweighs general quality signals for that user. Personalization can capture this dynamic.
- Member Recommendation Adds Expertise Discovery — Beyond content, the social graph reveals expertise. Users searching a topic benefit from seeing connections who are knowledgeable about it.
- Privacy Boundaries Must Be Respected — Social personalization involves sensitive cross-user signals. Privacy settings, consent, and disclosure must be first-class.
- Integration Must Stay Bounded — Social signals modulate ranking but cannot override quality. Bounded influence prevents social-graph manipulation from dominating rankings.
Innovation
How The System Works
The system reads the user's social graph, identifies connections who engaged with content relevant to the user's query, surfaces that content with social-context indicators, recommends socially-connected experts as related-people results, and respects privacy boundaries throughout.
- Read User Social Graph — Per user, retrieve the social graph (connections, follows, group memberships) with appropriate privacy consents. Output is the personalization graph for this user.
- Identify Connection Engagement — Per connection, identify content they have engaged with: shared, endorsed, commented. Engagement signals form the candidate-content pool.
- Filter By Query Relevance — Per query, filter connection-engaged content for topical relevance. Only relevant connection-engaged content surfaces.
- Score With Social Boost — Per candidate content, score combines query relevance with social-engagement strength. Multi-connection engagement boosts more than single-connection.
- Identify Expert Connections — Per query, identify connections with topical expertise. Their profiles surface as related-people results alongside content.
- Surface With Social Context — Personalized results render with social-context indicators ('shared by X', 'X is an expert in this topic'). Transparency lets the user evaluate.
- Capture Feedback — User reactions to socially-personalized results refine the model. Successful patterns strengthen; failed ones weaken.
Social Graph As Personalization Lens
The patent's load-bearing idea is to read the user's social graph as a lens that shapes search results. Friend engagement boosts content; expert connections surface as related people.
Trusted Connections Carry Personalized Authority
General authority signals are universal; trusted connections offer per-user authority. A friend's endorsement carries personalized weight that no global signal captures.
- Connection Engagement Signal — Per user, connections who engaged with content provide personalized relevance signal. The signal is precise per-user.
- Expert Connection Identification — Beyond content, expert connections surface as related-people results. Discovery extends beyond documents.
- Bounded Influence — Social signal modulates ranking but cannot override quality. The system stays robust to social-graph manipulation.
Technical Foundation
Technical Foundation
The patent specifies the social graph reader, the engagement aggregator, the topical relevance filter, the social-boost scorer, the expert identification model, and the privacy-bounded integration.
- Social Graph Reader — Per user, fetches the social graph respecting privacy consents. Graph includes connections, follows, group memberships, with relationship-strength signals.
- Engagement Aggregator — Per connection, aggregates engagement signals: shares, likes, comments, time spent. Aggregation is privacy-bounded.
- Topical Relevance Filter — Per query, filters connection-engaged content for topical relevance. Off-topic engagement does not contribute to query results.
- Social Boost Scorer — Per content candidate, combines query relevance with social engagement strength. Multiple connections engaging boosts more than single.
- Expert Identification Model — Per topic, identifies expert connections via expertise signals (publication history, prior engagement, declared expertise). Experts surface as related-people results.
- Privacy-Bounded Integration — Privacy settings filter who can see whose engagement signals. Default conservative; user controls expand scope where consent is given.
The Process
The Process
Social personalization runs in the query path for logged-in users with social-connected accounts. Latency stays within the standard query budget.
- Query From Logged-In User — Authenticated query arrives. User social-graph lookup activates.
- Fetch Social Graph — Per user, fetch the social graph with privacy filters applied. Output is the personalization context.
- Identify Connection Engagement — For the query topic, identify connection-engaged content. Engagement signals provide candidate content.
- Filter And Score — Topical relevance filter excludes off-topic; social boost scorer combines with query relevance. Output is ranked candidate content.
- Identify Expert Connections — Per query, expert identification produces related-people candidates from the social graph.
- Compose SERP — Standard SERP plus social-context indicators plus related-people block. Each personalized element labels transparently.
- Log And Learn — Engagement on personalized elements logs. Feedback refines the social-boost and expert-identification models.
Quality Control
Quality Control
Bad social personalization breaks trust. The patent specifies safeguards.
- Privacy Default Conservatism — Social signals require explicit consent. Default is conservative; users opt into broader social personalization.
- Bounded Influence — Social boost is bounded. Strong quality signals can still rank above socially-endorsed content if quality differs sharply.
- Manipulation Detection — Coordinated social engagement (paid likes, bot networks) is detected and discounted. Manipulation does not inflate rankings.
- Transparency Of Personalization — Personalized results label as such ('shared by X'). Users see why content surfaces and can evaluate.
- User Override — Users can disable social personalization or pause specific connection-based signals. First-class controls.
Real-World Application
Social personalization primitives ran in Google+ era products and influenced modern personalization features. The pattern of connection-engagement-boosted results plus expert-related-people is foundational to identity-aware search.
- Per-user Personalization Scope — Social graph per user produces per-user personalization. Different users see different personalized layers on the same query.
- Bounded Influence Magnitude — Social signal modulates but does not override. Quality signals retain primary ranking influence.
- Transparent User-Facing Label — Personalized results label transparently. Users see why social-personalized elements appear.
Why Identity-Aware Personalization Compounds
When users log in and the system reads their identity-linked social graph, personalization produces more relevant results. Sites earning real audience engagement (rather than synthetic metrics) win in identity-aware personalization layers.
Why Expert Discovery Reshapes Author Visibility
When the system surfaces expert connections, individual authors gain visibility tied to their established expertise rather than purely to their content's link profile. Personal-brand investment compounds in personalization.
<\/section>What This Means for SEO
What This Means for SEO
This patent personalizes results using social-graph signals (connection endorsements, expert identification), with bounded influence and manipulation detection. SEO implication: real audience engagement and genuine author expertise gain personalized visibility, while coordinated or synthetic social signals are discounted.
- Real Audience Engagement Wins Personalization — Connection-engagement signals boost content for users whose trusted contacts genuinely engaged. Earning a real audience that shares and endorses your content compounds in identity-aware personalization layers.
- Synthetic Social Metrics Are Discounted — Coordinated engagement like paid likes and bot networks is detected and discounted. Buying social signals does not inflate ranking, so invest in authentic engagement.
- Author Expertise Gains Visibility — The system surfaces expert connections as related-people results based on established expertise. Building a credible personal brand and demonstrable expertise earns visibility tied to the author, not just the content's links.
- Social Signal Is Bounded — Social boost modulates but cannot override quality, so strong quality content still ranks above socially-endorsed but weaker content. Personalization complements quality; it does not replace it.
- Personalization Is Per User — Different users see different personalized layers on the same query based on their own graphs. There is no single ranking to game for everyone; relevance to real communities is what scales.
- Transparency Lets Users Judge — Personalized results are labeled, so users see why content appears and can evaluate it. Manufactured social proof is exposed rather than hidden, reducing its value.
- Engagement Feedback Refines The Model — User reactions to personalized elements train the social-boost and expert models. Content that genuinely resonates with a community strengthens its personalized standing over time.