Selects the best textual representation for entity attributes. Drives how Knowledge Panel and entity cards render attribute values — the difference between '2.1 km' and 'just over a mile'.
Patent Overview
- Inventor
- Yossi Matias, others
- Assignee
- Google LLC
- Filed
- 2017
- Granted
- 2020-06-16
The Challenge
The Challenge
Entity attributes have multiple valid textual representations. '2.1 kilometers' or 'just over a mile' or '~2 km'. Per user, per context, the best representation differs. Selecting the right one is structural for usability.
- Attributes Have Multiple Valid Forms — Distances in km vs miles, dates in different formats, numbers with different precision. All valid; only some are best per context.
- User Locale Matters — Per user locale, unit and format conventions differ. Wrong locale produces friction.
- Context Determines Precision — Per context, precision needs differ. '2.1 km' for navigation; 'about 2 km' for general info.
- Templates Must Generalize — Per attribute type, template selection must work across many specific values.
- Selection Must Be Fast — Per Knowledge Panel render, attribute selection fits within latency budget.
Innovation
How The System Works
The system identifies entity attribute values, generates candidate textual representations per attribute, scores candidates by user-locale fit and context-precision fit, selects the best per (user, context), and renders into Knowledge Panel UI.
- Identify Entity Attribute Value — Per entity, identify attribute values to surface.
- Generate Candidate Representations — Per attribute, generate candidate textual representations (units, precision, formatting variations).
- Apply User Locale — Per user, locale-appropriate candidates favored.
- Apply Context Precision — Per context, precision-appropriate candidates favored.
- Score Candidates — Combined locale plus context plus aesthetic scoring.
- Select Best Representation — Top candidate selected for rendering.
- Render Into Panel UI — Selected representation rendered.
Right Representation For Right Context
The patent's load-bearing idea is that attribute values have multiple valid representations; selection per (user, context) determines usability. The selection layer is the structural primitive for friction-free entity surfaces.
Locale Plus Context Drives Selection
Per user locale, conventions differ. Per context, precision differs. The combined signal selects the best representation.
- Candidate Generation — Per attribute, multiple candidate representations generated.
- Locale Fit — Per user locale, locale-appropriate candidates favored.
- Context Precision Fit — Per context, precision-appropriate candidates favored.
Technical Foundation
Technical Foundation
The patent specifies the attribute identifier, candidate generator, locale matcher, context matcher, scorer, selector, and renderer.
- Attribute Identifier — Per entity, identifies attribute values.
- Candidate Generator — Per attribute, generates candidate representations.
- Locale Matcher — Per user locale, locale-fit scored.
- Context Matcher — Per context, precision-fit scored.
- Scorer — Combined locale plus context scoring.
- Renderer — Selected representation rendered.
The Process
The Process
Per Knowledge Panel render, attribute selection runs at query time.
- Panel Triggered — Per query, Panel surfaces.
- Identify Attributes — Per entity, attributes identified.
- Generate Candidates — Per attribute, candidates generated.
- Apply Locale — Locale fit scored.
- Apply Context — Context fit scored.
- Select — Top candidate selected.
- Render — Selected representation rendered into Panel UI.
Quality Control
Quality Control
Wrong representations produce friction. The patent specifies safeguards.
- Locale-Validation — Per locale, candidate appropriateness validated against labeled data.
- Context-Precision Calibration — Per context type, precision conventions calibrated.
- Aesthetic Consistency — Per surface format, aesthetic consistency enforced.
- Adversarial Defense — Manipulated entity-attribute data filtered before representation selection.
- Continuous Recalibration — Locale, context, aesthetic models recalibrate against fresh data.
Real-World Application
Textual attribute selection underpins Knowledge Panel rendering at scale. The pattern of locale-plus-context-driven selection makes entity surfaces feel native across user contexts.
- Multi-candidate Selection Pool — Per attribute, multiple candidate representations generated.
- Locale-aware Personalization — Per user locale, locale-appropriate candidates favored.
- Context-aware Precision Adaptation — Per context, precision-appropriate candidates favored.
Why Multi-Unit Attribute Markup Helps Knowledge Panel Coverage
Pages exposing entity attributes in multiple units (metric and imperial, multiple date formats) provide the candidate-generation layer with more options. Knowledge Panel renders the best per user locale.
Why Aesthetic Format Choices Matter
Per attribute, aesthetic consistency matters. Pages with consistent attribute formatting align with selection conventions and contribute clean candidates.
<\/section>What This Means for SEO
What This Means for SEO
This patent selects the best textual representation of an entity attribute for the user's locale and context, choosing among units, precision, and formatting. SEO implication: exposing attributes in multiple units and consistent formats gives the system clean candidates to render natively per user.
- Expose Attributes In Multiple Units — The candidate generator picks among representations like metric and imperial or multiple date formats. Pages that provide attributes in multiple units give the selector more options to render the best fit per user locale.
- Consistent Formatting Produces Clean Candidates — Aesthetic consistency is enforced in selection. Pages with consistent attribute formatting align with selection conventions and contribute clean candidates, where erratic formatting produces noisy ones.
- Locale Drives Representation Choice — Per user locale, unit and format conventions differ and locale-appropriate candidates are favored. Providing locale-friendly representations of your data helps it render natively rather than awkwardly converted.
- Context Sets The Precision — Precision needs vary by context, exact for navigation and approximate for general info. Offering attribute values at appropriate precision lets the system pick the right level rather than forcing an over- or under-precise form.
- Multi-Unit Markup Widens Panel Coverage — More candidate representations mean the Knowledge Panel can render correctly across more user contexts. Multi-unit, multi-format attribute data broadens the contexts in which your entity surfaces cleanly.
- Manipulated Attribute Data Is Filtered — Manipulated entity-attribute data is filtered before representation selection. Accurate, genuine attribute values are the input the system works from; fabricated values get removed.
- Clean Structured Data Is The Foundation — The whole selection layer runs on identifiable attribute values. Exposing entity attributes through clean structured data is what makes them available for native, context-appropriate rendering at all.