Per user, models domain expertise across topics. The user-side authority/expertise model — complement to Google's document-side authority work, with applications across enterprise search, social search, and personalization.
Patent Overview
- Inventor
- Susan T. Dumais, others
- Assignee
- Microsoft Corporation
- Filed
- 2007
- Granted
- 2012-02-21
The Challenge
The Challenge
Per user, domain expertise varies across topics. Modeling per-user per-topic expertise enables expert finding, content personalization (advanced vs introductory), and authority-weighted social search. The user-side authority model complements document-side authority work.
- User Expertise Varies By Topic — Per user, expertise differs across topics.
- Expertise Has Multi-Source Signal — Per (user, topic), content engagement, content creation, social signals combine.
- Expert-Finding Enables Search Applications — Per topic, identifying experts enables expert-routing and authority-weighted retrieval.
- Personalization Depth Depends On Expertise — Per user, expertise level shapes appropriate content depth.
- Privacy Preservation Required — Per user, expertise signals handled with privacy.
Innovation
How The System Works
The system captures multi-source expertise signals per user per topic, infers per-(user, topic) expertise level, applies expertise model to expert-finding, personalization, and authority-weighted retrieval, and respects privacy.
- Capture Expertise Signals — Per user, content engagement, creation, social signals captured.
- Per-Topic Expertise Inference — Per (user, topic), expertise inferred.
- Build User Expertise Profile — Per user, per-topic expertise profile built.
- Apply To Expert-Finding — Per topic, top experts identified.
- Apply To Personalization — Per (user, query), expertise level shapes content depth.
- Apply To Authority-Weighted Retrieval — Per (user, query), expert-authored content weighted appropriately.
- Privacy Preserve — Per user, signals handled with privacy.
User-Side Authority Model
The patent's load-bearing idea is that user expertise is a measurable property complementing document authority. Per-(user, topic) expertise enables expert-finding, depth-appropriate personalization, and authority-weighted retrieval.
Per-Topic Expertise
Per user, expertise differs across topics. Per-topic modeling captures the granularity.
- Multi-Source Expertise Signals — Engagement, creation, social signals combine.
- Per-Topic Inference — Per (user, topic), expertise inferred.
- Multi-Application — Expert-finding, personalization, authority-weighting all consume model.
Technical Foundation
Technical Foundation
The patent specifies the signal capturer, per-topic inferrer, profile builder, expert finder, personalizer, authority weighter, and privacy layer.
- Signal Capturer — Per user, multi-source signals captured.
- Per-Topic Inferrer — Per (user, topic), expertise inferred.
- Profile Builder — Per user, expertise profile built.
- Expert Finder — Per topic, experts identified.
- Personalizer — Per (user, query), depth-appropriate personalization.
- Authority Weighter — Per (user, query), expert-authored content weighted.
The Process
The Process
Expertise inference runs continuously; applications run per query.
- Capture Signals — Per user, signals captured.
- Infer Per Topic — Per (user, topic), expertise inferred.
- Build Profile — Per user, profile built.
- Receive Query — Query arrives.
- Apply Expert-Finding — Top experts identified.
- Apply Personalization — Depth-appropriate personalization.
- Apply Authority-Weighting — Expert-authored content weighted.
Quality Control
Quality Control
Wrong expertise inference damages applications. The patent specifies safeguards.
- Multi-Signal Convergence — Per (user, topic), expertise flag requires multi-signal convergence.
- Privacy Preservation — Per user, signals handled with privacy.
- Engagement Validation — Per (user, topic), engagement validates inference.
- Bias Detection — Per profile, bias patterns detected.
- Continuous Recalibration — Models refresh.
Real-World Application
Domain-expertise modeling underpins expert-finding in enterprise search, depth-appropriate personalization, and authority-weighted social search. The user-side authority pattern complements document-side authority.
- Per-(user, topic) Granularity — Each user-topic pair gets expertise inference.
- Multi-source Signal Combination — Engagement, creation, social signals combine.
- Multi-application Application Scope — Expert-finding, personalization, authority-weighting.
Why Demonstrated Expertise Compounds In Modern Search
Per (user, topic), demonstrated expertise (content creation, engagement, citation) compounds. Building visible expertise across a topic compounds across multiple application surfaces.
Why Author-Level Signals Matter
Per author, expertise per topic accumulates over time. Author-level expertise signals are increasingly significant in modern search beyond per-page signals alone.
<\/section>What This Means for SEO
What This Means for SEO
The system models per-user, per-topic expertise from content creation, engagement, and social signals, complementing document-side authority with a user-side authority model. SEO implication: author-level demonstrated expertise compounds across a topic and increasingly matters beyond per-page signals.
- Author-Level Signals Accumulate — Expertise is modeled per author per topic and builds over time. Consistent authorship in a defined topic accumulates a user-side authority signal that individual pages alone do not carry. Attribute and concentrate your authors' work.
- Demonstrated Expertise Beats Claimed Expertise — The model reads content creation, engagement, and citation, not bio claims. Visible, engaged-with, cited contributions in a topic establish expertise the system can measure. Show the work.
- Topic Focus Concentrates The Signal — Expertise is per-topic, so spreading an author thinly across unrelated subjects dilutes each topic signal. Authors who go deep in a focused domain build a stronger, more legible expertise profile.
- Engagement With Your Content Feeds Expertise — Content engagement is an expertise input. Material that earns genuine reader interaction reinforces the author's modeled expertise in that topic, linking content quality to author authority.
- Depth-Appropriate Content Matches Reader Expertise — The model also informs depth personalization, matching introductory versus advanced content to user expertise. Offering both entry-level and advanced material lets your content fit readers at every expertise level rather than missing one end.
- Expert-Finding Surfaces Recognized Authors — The system enables expert-routing and authority-weighted retrieval. Authors recognized as topic experts are more likely to be surfaced for that topic, so building a named expert profile is a discovery investment.
- Cross-Surface Expertise Compounds — User-side authority applies across enterprise search, social search, and personalization. Expertise demonstrated in one surface reinforces the author's standing across the others, so a coherent expert identity pays off broadly.