Generates data facets (dimensions) over a dataset based on data types of its elements, then recommends the facets most useful to surface for a given query, supporting faceted-search and explore-style result presentation.
Patent Overview
- Inventor
- Nitin Gupta
- Assignee
- Google LLC
- Filed
- 2022-06-08
- Granted
- 2023-12-14 (published application)
- Application Number
- US 17/834,950
The Challenge
Datasets Need Useful Slices, Not Just Ranked Lists
When a search returns a large set of results from a dataset, ranked listing is insufficient. Users want to slice the data by meaningful dimensions: price ranges, categories, geography, time period, attribute values. The system needs to generate facets automatically from the dataset's structure and recommend which facets to surface based on what would be most useful for a given query.
- Faceted Search Beats Ranked Lists For Exploration — For exploratory queries with many candidate results, slicing by facets is more useful than ranking. Ranked lists answer 'which is best'; facets answer 'what are my options'.
- Manual Facet Definition Doesn't Scale — Hand-authoring facets per dataset works for a few well-known datasets. It does not work at the scale of every searchable dataset on the web.
- Need Automatic Generation From Data Types — The data types of elements (categorical, numeric, geographic, temporal) imply what facets make sense. Generation from data types is the scalable approach.
- Recommendation Must Be Query-Aware — Not every facet is useful for every query. The system needs to recommend which facets to surface based on the current query's intent.
Innovation
Type-Driven Facet Generation Plus Query-Aware Recommendation
The method receives data associated with a dataset where each data entry has elements with one or more data types. It generates one or more data facets for each entry based on the associated data type, then produces transformations and recommendations of which facets to surface. Categorical types generate category facets; numeric types generate range facets; geographic types generate region facets.
- Receive Dataset — Data associated with a dataset arrives. Entries have elements, each with a known or inferable data type.
- Identify Data Types — For each element, determine its data type: categorical, numeric, boolean, geographic, temporal, etc.
- Generate Type-Appropriate Facets — For each (entry, element, type) combination, generate one or more facets. Categorical types yield category facets; numeric types yield range buckets; temporal types yield date buckets.
- Compute Transformations — Generate transformations that map raw values to facet values (e.g., raw prices to price ranges, raw dates to month buckets).
- Receive Query Context — On a query, take the query into account when deciding which facets to recommend.
- Recommend Facets For Surface — Surface the facets most likely to be useful for the query: those that meaningfully reduce the result set, that align with the query's expected refinement axes, and that produce well-balanced buckets.
- Render Faceted Result Surface — User sees the result set with recommended facets exposed for slicing. Each facet click filters the results dynamically.
Facets From Types, Recommendation From Query
The patent separates facet generation (from data types) from facet recommendation (from query intent). The separation is what makes the approach scale: facet generation is offline and type-driven; recommendation is online and query-aware.
Two Stages: Generate Offline, Recommend Online
All possible facets are generated offline from the dataset's data types. The query-aware recommendation runs online to pick the right subset.
- Data Types Drive Generation — Each element's data type implies the facet shape: categorical to categories, numeric to ranges, geographic to regions, temporal to dates.
- Query Drives Recommendation — Online, the query context selects which generated facets to surface. Different queries surface different facet subsets from the same dataset.
What This Means for SEO
What This Means for SEO
Faceted search surfaces shape how users explore result sets. Knowing the type-to-facet mapping and query-aware recommendation informs how to structure content datasets for faceted discovery.
- Structured Data Yields Better Facets — When your content datasets have clean type annotations (categorical, numeric, geographic, temporal), the facet generation works better. Schema-rich product catalogs, structured event listings, and well-typed databases produce richer facets.
- Type Diversity Multiplies Facet Coverage — Datasets with multiple data types (text + price + date + location) generate facets across all dimensions. Single-type datasets get fewer facets and less faceted exploration.
- Range And Bucket Strategy Matters — Numeric ranges and date buckets need to be useful for typical queries. Buckets that are too narrow or too wide produce poor faceted experiences. Anticipate the buckets your audience actually wants.
- Facets Compound Discoverability — Each facet click is an entry point to a slice of your content. Pages that participate in multiple facets get exposed across multiple exploration paths.