Identifies query templates composed of terms plus members of entity categories, then ranks the templates so valid templates can produce concrete query suggestions for the user.
Patent Overview
- Inventor
- Nitin Gupta
- Assignee
- Google LLC
- Filed
- 2014-06-03
- Granted
- 2016-12-27
- Application Number
- US 14/295,153
The Challenge
Suggestion Systems Need Templates Beyond Literal History
Suggestion systems built on literal query history struggle with novel intents. If a user has not searched a specific phrase before, history-based suggestions cannot help. The system needs templates that capture the shape of common intents and can be instantiated with entity members to produce concrete suggestions for any query, including ones never seen before.
- Literal History Misses Long-Tail Intent — Most queries are issued few or no times in literal history. Suggestion systems that depend on literal recurrence fail on the long tail.
- Templates Generalize Across Entities — A template like '[movie] showtimes' instantiates with any movie entity. One template produces millions of concrete suggestions across the entity category.
- Template Quality Varies — Some templates are widely useful (most users want '[movie] showtimes'). Others are too narrow or unnatural. The system needs to rank templates so only valid ones drive suggestions.
- Need Entity-Category Awareness — Templates reference members of entity categories (movies, restaurants, products). The system must know which categories to anchor templates against and which entity members are valid fillers.
- Validity Is Tunable — What counts as a valid template depends on the count of supporting occurrences. The ranking surface is what determines validity, not a hand-coded rule.
Innovation
Templates Plus Entity Members Plus Ranking
The system identifies query templates that include one or more terms and a member of an entity category. Each template is ranked using a count of occurrences of the template instantiations, with higher-ranked templates designated as valid. Valid templates drive query suggestions for users by instantiating against the user's current entity context.
- Mine Templates From Query Logs — Scan historical queries for patterns where one or more terms are paired with members of known entity categories. Each pattern becomes a candidate template.
- Identify Entity-Category Membership — For each candidate template, identify which entity category the variable position belongs to (movies, restaurants, products, etc.).
- Count Instantiations — For each template, count how many times the template has been instantiated in the query log. Each instantiation is a concrete query that fits the template's shape.
- Rank Templates — Order templates by their instantiation count, with high-count templates ranked as more valid. Other quality signals (CTR on resulting queries) feed the ranking too.
- Mark Validity — Templates above a threshold rank are marked valid for use in suggestion generation. Below-threshold templates are dropped.
- Generate Suggestions — When a user query arrives, identify the user's current entity context. Apply valid templates whose entity category matches the context. Each application produces a concrete query suggestion.
- Surface Top Suggestions — Rank the generated suggestions by template rank and entity-context fit. Surface the top suggestions in the autocomplete or related-search interface.
Template Plus Entity Generalizes Beyond History
The patent moves suggestion generation from history-only to template-driven. Templates capture the shape of common intents; entity members provide the concrete fillers. The combination produces suggestions even for entity-context pairs never seen before in raw history.
Shape Plus Entity Equals Suggestion
Templates encode the shape; entity-category members encode the specifics. Apply a template to an entity context and you have a concrete suggestion.
- Template Shape — A pattern with one or more terms and a variable position for an entity-category member. Captures the syntactic structure of an intent.
- Entity Members — Concrete fillers for the variable position. Drawn from known entity collections so the resulting suggestions are grounded in real entities.
- Ranking And Validity — Templates earn validity through instantiation count and downstream quality. Only valid templates produce suggestions.
Technical Foundation
Template Structure And Ranking
The framework defines templates explicitly with variable slots and ranks them based on observed usage.
- Query Template — A pattern with terms plus one or more entity-category slot positions. Templates are reusable across many concrete instantiations.
- Entity Category — The category of entities that fills the slot position (movies, products, restaurants, books, etc.). Membership is determined by the entity index.
- Instantiation Count — Number of times the template has appeared in the query log filled with a member of the entity category. Drives the template's rank.
- Validity Threshold — Minimum rank at which a template is treated as valid for suggestion generation. Templates below threshold do not produce suggestions.
Key Insight: Treating suggestions as template-plus-entity rather than as literal query strings is what makes the long-tail addressable. A single template instantiated against thousands of entity members produces thousands of suggestions, most of which were never literally observed in history. The template abstraction is the scalability layer.
<\/section>What This Means for SEO
What This Means for SEO
Query suggestion templates shape which queries users actually issue. Understanding the template system informs how to target query shapes rather than specific phrasings.
- Target Template Shapes, Not Just Phrases — When you optimize for a query, target the template shape (entity + qualifier + intent verb), not just the exact phrasing. The template generates many concrete suggestions; covering the shape captures them all.
- Entity Membership Is The Eligibility Gate — Pages that are clearly about a recognized entity member of a category participate in all templates that bind to that category. Strong entity association multiplies query-shape coverage.
- Common Templates Drive The Long Tail — The high-instantiation templates ('[movie] showtimes', '[product] reviews', '[city] weather') drive enormous query volume. Content built around these template shapes serves a long tail of concrete instantiations.
- Niche Entity Membership Matters — If your content covers an entity that's a member of a high-volume template's category, you get suggestion exposure for every template instantiation. Niche entities in high-volume categories are an asymmetric opportunity.
- Schema Markup Helps Template Matching — Templates bind to entity categories. Explicit schema markup (Product, Movie, LocalBusiness, etc.) helps the system recognize which categories your content's entities belong to, which determines which templates select your content.