Earlier search augmentation work. Identifies augmentation queries from user interactions or machine generation, stores them, and uses geographic location and language signals to select augmentation queries that improve retrieval for incoming user queries.
Patent Overview
- Inventor
- Anand Shukla
- Assignee
- Google LLC
- Filed
- 2008-05-16
- Granted
- 2013-01-01
- Application Number
- US 12/121,973
The Challenge
Search Needs Augmentation Beyond Literal Terms
When users submit search queries, the literal terms often miss the broader intent. Augmenting the query with related terms — derived from past user interactions, geographic context, or language — improves retrieval. The system needs an augmentation framework that draws from multiple sources and selects relevant augmentations for each incoming query.
- Literal Queries Are Often Underspecified — Users type brief queries that miss the surrounding context the search engine could use to improve retrieval.
- User Interactions And Machine Generation Both Inform Augmentation — Augmentation queries can come from observed user reformulations or be generated by the system. Both sources contribute to the augmentation query data store.
- Geography And Language Are Contextual Selectors — When choosing which augmentation queries to apply, the user's geographic location and language are powerful filters. Location-specific augmentations apply to users in that location; language-specific augmentations apply to that language.
Innovation
Multi-Source Augmentation Plus Context-Aware Selection
Augmentation queries are identified from user interactions or generated by the system. They are stored in an augmentation query data store. When a user submits a search query, the query's terms plus optional additional information (geographic location, language) are used to identify stored augmentation queries and select one or more similar ones. The selected augmentations are used by the search engine to perform retrieval.
- Identify Augmentation Queries — Augmentation queries come from two sources: observed user interactions (reformulations, clicks, follow-up queries) and machine generation (synthesized from intent models).
- Store In Data Store — Augmentation queries are persisted in a dedicated data store keyed for fast lookup by intent or topic.
- Receive User Query Plus Context — User submits a query. Context information (geographic location, language) is captured alongside the query terms.
- Select Matching Augmentations — Look up augmentation queries that are similar to the user's query, filtered by geographic location and language. Multiple augmentations can match.
- Apply Augmentations To Retrieval — The search engine uses the selected augmentation queries to perform retrieval. Documents matching any of the augmentation queries become candidates.
What This Means for SEO
What This Means for SEO
Search augmentation expands the queries that retrieval considers, often beyond the user's literal terms. Knowing the mechanism informs how to think about geographic and language-specific content.
- Geographic Context Shapes Augmentation — Augmentations are filtered by user location. Content with strong geographic signals participates in location-filtered augmentations and reaches more queries.
- Language-Specific Augmentation — Language is a filter on augmentation selection. Content in the user's language is preferred for augmentation matching.
- User-Interaction-Derived Augmentation — Real user reformulations inform the augmentation store. Content that satisfies common reformulations gets compounding exposure as the augmentation pulls related queries toward it.
