Domain-Focused Q&A Model Fine-Tuning With Supporting Citations

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing search systems fail to adequately search across multiple formats of information sources, provide relevant context, and adapt language models to domain-specific data, leading to inadequate answers and lack of supporting citations for natural language questions.

Innovation Solution

A system and method that fine-tunes a pre-trained machine learning model using domain-specific files to generate natural language answers and provide supporting citations, incorporating retrieval and generation models to enhance domain-specific responses.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If existing search systems use keyword-based searching, then search speed is maintained, but the ability to understand natural language questions and provide contextually relevant answers deteriorates

Engineering Contradiction:

Improvenatural language query capabilityVSAvoidanswer accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The system changes the operational parameters of the search model by fine-tuning it on domain-specific question-answer pairs. This transforms the search model from a generic keyword-matching system to a domain-adapted system that understands natural language queries in specific domains, thereby improving both ease of operation and measurement precision

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system performs preliminary fine-tuning of the search model using domain-specific training data before deployment. This preliminary action prepares the model to handle domain-specific terminology and context, enabling it to accurately process natural language questions in the target domain without requiring complex runtime adjustments

Inventive Principle:

Principle #10Preliminary action

2Adaptability or versatility

If search systems search across multiple file formats, then information coverage is improved, but the ability to provide relevant context and source documents deteriorates

Engineering Contradiction:

Improvemulti-format search capabilityVSAvoidcontext relevance

Core Design Contradiction:

Adaptability or versatilityVSLoss of information

Solution Approach 1:

The system introduces an intermediary retrieval component that acts as a bridge between the multi-format file search and the final answer generation. This intermediary selectively retrieves and prioritizes relevant context and source documents from the search results, ensuring that information from multiple formats is processed in a way that preserves context relevance and prevents information loss

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If language models are not fine-tuned on domain-specific data, then model generality is maintained, but domain-specific vocabulary understanding and answer accuracy deteriorate

Engineering Contradiction:

Improvedomain-specific vocabulary understandingVSAvoidmodel fine-tuning process

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The system performs preliminary fine-tuning of the language model on domain-specific question-answer pairs before deployment. This preliminary action adapts the model to domain-specific vocabulary and contexts, enabling it to understand and generate accurate domain-specific answers without requiring complex runtime adjustments or interventions

Inventive Principle:

Principle #10Preliminary action

4Productivity

If existing search systems provide answers without supporting citations, then answer generation speed is improved, but the ability to provide verifiable and traceable information deteriorates

Engineering Contradiction:

Improveanswer generation speedVSAvoidanswer verifiability

Core Design Contradiction:

ProductivityVSReliability

Solution Approach 1:

The system implements a feedback mechanism where the generation model is trained to produce answers accompanied by supporting citations from the retrieved documents. This feedback loop ensures that the model learns to generate verifiable answers with proper attributions, maintaining answer generation speed while improving reliability and traceability of the information provided

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS12608341B2Domain-focused natural language question and answer

Publication Date: 2026.04.21 LOCKHEED MARTIN CORP

AI summary

A domain-focused natural language question and answer system. The system includes one or more electronic processors. The one or more electronic processors are configured to create, from a plurality of files, a plurality of fine-tuning data examples, each respective fine-tuning data example of the plurality of fine-tuning data examples including a question, an answer, and a supporting citation. The one or more electronic processors are further configured to fine-tune a pre-trained foundational machine learning model using the plurality of fine-tuning data examples to generate a fine-tuned machine learning model, input a natural language question to the fine-tuned machine learning model, and retrieve, from the fine-tuned machine learning model, a natural language answer to the natural language question and an indication of one or more files, one or more relevant sections, or both that the natural language answer is based on.