Generative Search Context Selection for Diverse Query Coverage

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current search systems generate long-form responses that lack diversity and completeness, often focusing on a single aspect of complex queries due to bias from top-ranked documents, leading to lower-quality answers with hallucinations.

Innovation Solution

Extract relevant portions from multiple top-ranked resources, ensuring diversity by selecting passages that maximize dissimilarity and relevance, and optionally incorporating related queries to enhance completeness.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If top-ranked documents are used as context for generative language model, then hallucinations are minimized, but response diversity and completeness deteriorate

Engineering Contradiction:

Improveaccuracy of long-form responseVSAvoiddiversity of aspects covered

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent segments the context selection process into multiple stages: first identifying top-ranked documents, then partitioning them into passages, then selecting most-relevant passages, and finally adding diverse passages from related queries. This segmentation allows the system to balance reliability (through top-ranked document selection) with diversity (through multi-stage diverse passage selection).

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent merges passages from multiple sources: top-ranked documents for the main query and top-scoring documents from related queries. By combining these diverse sources, the system maintains reliability from authoritative sources while achieving diversity through multiple document sources and query perspectives.

Inventive Principle:

Principle #5Merging (Combining)

2Loss of information

If multiple passages from top-ranked documents are selected, then completeness improves, but diversity deteriorates due to bias from similar queries

Engineering Contradiction:

Improvecompleteness of query coverageVSAvoiddiversity of passages

Core Design Contradiction:

Loss of informationVSAdaptability or versatility

Solution Approach 1:

Instead of only selecting from top-ranked documents for the main query, the patent inverts the approach by also selecting passages from related queries. This inversion introduces diversity by bringing in perspectives from different query angles while maintaining completeness through comprehensive passage selection.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The patent changes the selection parameters by introducing diversity constraints and completeness constraints as separate optimization objectives. The system adjusts passage selection based on multiple parameters including relevance score, diversity metrics, and coverage of different query aspects, rather than relying on a single ranking parameter.

Inventive Principle:

Principle #35Parameter changes

3Loss of information

If diverse passages are selected from related queries, then completeness improves, but system complexity increases

Engineering Contradiction:

Improvecompleteness of responseVSAvoidcomplexity of passage selection process

Core Design Contradiction:

Loss of informationVSDevice complexity

Solution Approach 1:

The patent performs preliminary actions by pre-identifying related queries and pre-selecting top-scoring documents for each related query before final passage selection. This preliminary organization of data structures and pre-computation of document scores reduces the complexity of the final diverse passage selection process.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary passage selection process that acts as a mediator between the main query and related queries. This intermediary layer organizes and filters passages from multiple sources before presenting them to the generative language model, simplifying the overall system architecture and making the complexity manageable through structured intermediate processing.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS20260072965A1Controlled content diversity in retrieval for generative search

Publication Date: 2026.03.12 GOOGLE LLC

AI summary

Implementations relate to techniques for accounting for diversity and/or completeness when generating a long-form natural language response for a search query. Implementations may identify the most relevant passage in a top-ranking documents for the query and then select, from among the most-relevant passages, those passages that meet inclusion criteria, e.g., a minimum relevance to the query, maximizing diversity with other relevant passages, etc. The passages (or portions thereof) that meet the inclusion criteria may be provided with the query to a generative language model, which generates a long-form response to the query. Some implementations may add additional passages to the potential pool of passages, the additional passages identified from top-scoring documents for queries related to the query provided by the user.