Foundation Model Content Summaries for Accurate Query Matching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing search engines use keyword matching to associate content with queries, which is costly, slow, and fails to understand the context of the content creator's goals, leading to keyword misinterpretation issues.

Innovation Solution

A system utilizing two foundation models to summarize and align content summaries with target characteristics defined by the content creator, ensuring the summary accurately reflects the creator's goals through an automated feedback loop.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If keyword matching is used to associate content with queries, then the search engine can process queries, but the process becomes costly and slow

Engineering Contradiction:

Improvequery processing speedVSAvoidcomputational cost

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The system performs preliminary actions by generating summaries of content instances and their associated characteristics before query processing. These pre-computed summaries and characteristics are stored and reused during query matching, eliminating the need for repeated full-content analysis and reducing both computational cost and processing time.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system extracts key characteristics from full content instances using foundation models, creating condensed representations that capture essential information. This extraction process separates the critical features from the complete content, enabling faster comparison and matching during query processing without analyzing entire content instances.

Inventive Principle:

Principle #2Taking out (Extraction)

2Productivity

If foundation models summarize content instances, then query matching becomes faster, but some important characteristics may be excluded from the summary

Engineering Contradiction:

Improvequery matching speedVSAvoidexcluded characteristics

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

The system implements feedback by having the second foundation model evaluate whether generated summaries align with target characteristics. When misalignment is detected, the system iteratively refines the summary generation process, using the evaluation feedback to adjust and improve summary quality, ensuring important characteristics are retained.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system changes parameters by adjusting summary generation settings based on target characteristics. The foundation model modifies summary properties such as length, detail level, and focus areas to better align with the content creator's goals while maintaining query matching efficiency.

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If the system uses multiple foundation models with feedback loops to ensure summary alignment, then summary accuracy improves, but system complexity increases

Engineering Contradiction:

Improvesummary alignment accuracyVSAvoidsystem architecture

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system applies universality by using foundation models that perform multiple functions: the first foundation model generates summaries, the second evaluates alignment with target characteristics, and both contribute to the iterative refinement process. This multi-functional approach reduces the need for separate specialized components for each task.

Inventive Principle:

Principle #6Universality (Multi-functionality)

4Productivity

If summaries are limited to a threshold size, then processing efficiency improves, but important characteristics may be omitted

Engineering Contradiction:

Improveprocessing efficiencyVSAvoidomitted characteristics

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

The system dynamically adjusts summary parameters including size constraints based on the importance and density of characteristics. The foundation model optimizes the balance between summary length and information completeness, modifying parameters such as token limits, sentence counts, and compression ratios to retain essential characteristics within efficiency thresholds.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12579203B2Automatic generation of content for query matching in foundation model-based content providing systems

Publication Date: 2026.03.17 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

The disclosed system optimizes, via the use of foundation models such as large language models (LLMs), the manner in which a search engine associates instances of content (e.g., webpages) with a query. The system employs a first foundation model to summarize an instance of content. The system the employs a second foundation model to compare the summary of the instance of content generated by the first foundation model to target characteristics defined by a content creator. If the comparison confirms that the summary of the instance of aligns with the target characteristics, the system provides the summary to a search engine for query matching purposes. If the comparison denies that the summary of the instance of aligns with the target characteristics, the system employs the first foundation model to re-summarize the instance of content using feedback received from the second foundation model.